Indonesian J our nal of Electrical Engineering and Computer Science V ol. 42, No. 1, April 2026, pp. 194 ∼ 204 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v42.i1.pp194-204 ❒ 194 A graph neural netw ork framew ork f or v ascular str eak dieback r ecognition Slamin 1 , Rizk y Alfanio Atmok o 1 , Antonius Cah ya Prihandok o 1 , Muhammad Ariful Fur qon 1 , Qurr ota A ’yuni Ar Ruhimat 1 , Annisa Fitri Maghir oh Har vyanti 1 , Bayu T aruna W idjaja Putra 1 , Roslan Hasni 2 1 Department of Computer Science, F aculty of Computer Science, Uni v ersity of Jember , Jember , Indonesia 2 F aculty of Computer Science and Mathematics, Uni v ersiti Malaysia T erengg anu, K uala Nerus T erengg anu, Malaysia Article Inf o Article history: Recei v ed Dec 15, 2025 Re vised Jan 31, 2026 Accepted Mar 4, 2026 K eyw ords: Agricultural image analysis Cocoa leaf disease Graph neural netw ork Hybrid CNN–GNN V ascular streak dieback ABSTRA CT V ascular streak dieback (VSD) is one of the most destructi v e diseases af fecting cocoa production in Southeast Asia, including Indonesia, where early visual symptoms are often subtle and spatially distrib uted across the leaf sur - f ace. Con v entional image-based disease recognition approaches, particularly those relying solely on con v olutional neural netw orks (CNNs), are ef fecti v e in e xtracting local visual features b ut remain limited in modeling long-range structural relationships such as v enation disruption and lesion spread. T o ad- dress this limitation, this study in v estig ates a h ybrid CNN-graph neural netw ork (CNN-GNN) frame w ork for automated VSD recognition from cocoa leaf im- ages. A primary dataset consisting of 1,000 RGB images collected directly from cocoa plantations in Jember Re genc y w as used to reect realistic eld condi- tions. In the proposed approach, CNNs are emplo yed for local feature e xtraction, while graph-based representations enable GNNs to capture global relational pat- terns through message passing. Experimental results demonstrate stable learning beha vior and strong classication performance, achie ving a maximum v alidation accurac y of 95.2% and an area under the curv e (A UC) of approximately 0.94. Further analysis sho ws balanced precision and recall across classes, indicating reliable discrimination between Sehat and VSD-infected lea v es. These ndings suggest that h ybrid CNN-GNN modeling pro vides an ef fecti v e strate gy for cap- turing both local and distrib uted structural characteristics of VSD symptoms and highlights the potential of graph-based reasoning to complement con v olutional feature learning in plant disease diagnostics. This is an open access article under the CC BY -SA license . Corresponding A uthor: Slamin Department of Computer Science, F aculty of Computer Science Uni v ersity of Jember , Jember , Indonesia Email: slamin@unej.ac.id 1. INTR ODUCTION Cocoa is an essential product for f armers throughout Indonesia, including those in Jember . Ho we v er , this does not mean that cocoa f arming is without its challenges. Indonesia, on the other hand, is considered geographically suitable for cocoa culti v ation, b ut geograph y alone does not determine the success of cocoa culti v ation. One cause of cocoa crop f ailure is fung al disease, one of which is v ascular streak dieback (VSD), caused by Ceratobasidium theobromae, and is considered a serious threat [1]. VSD in cocoa causes anatomical and ph ysiological symptoms that, if left untreated, will lead to decreased yields and will af fect the long-term results of cocoa plantations. When production decreases b ut demand remains constant or increases, this will J ournal homepage: http://ijeecs.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 195 cause cocoa prices to rise [2]. Se v eral studies, such as those from [2], sho w that cases of VSD in cocoa sho w a pattern of increasing in Southeast Asia, including Indonesia. Based on [3], Indonesia itself is one of the countries that produces cocoa and e xports cocoa in signicant quantities, which is supported by the latest report from BPS [4]. Cocoa diseases can be visually recognized, one of which is from the color of the lea v es. This is supported by research results from [1], [2], [5], which stated that there are color changes in the lea v es of cocoa plants e xperiencing VSD disease symptoms. Con v olutional n e ural netw orks (CNNs) ha v e been widely used in plant disease recognition due to t heir ability to e xtract local te xture and color features from leaf imagery [6]. Pre vious researches such as Mohanty et al. [7] sho wed that deep CNNs could classify plant diseases from RGB images, meanwhile T oo et al . [8] com- pared multiple transfer learning architectures for agricultural applications, achie ving signicant classication accurac y impro v ements. More recently , K ou a ssi et al. [9] applied CNNs detecting diseases of cocoa leaf, where CNN-based feature e xtraction combined with XGBoost classication yielded reliable results. Ho we v er , CNNs persist limitation in modeling long-range spatial dependencies and structural relationships acros s dif ferent leaf re gions. T o o v ercome these constraints, researchers ha v e e xplored graph neural netw orks (GNNs), which model e xplicitly relational structures and interactions among image re gions [10]-[17]. F or e xample, Zhao et al. [18] recommended a structural graph learning frame w ork for image classication, P ark et al. [19] were utilizing graph instruments to goal about visual relationships, and W u et al. [20] established adapti v e graph con v o- lutional netw orks for ne-grained visual recognition. Although these GNN-based models ef fecti v ely capture spatial dependencies, the y often lack comprehensi v e visual conte xt compared to CNNs. Accordingly , recent studies inte grated both paradigms through CNN-GNN h ybrid architectures [21]-[26], combination of CNN’ s strength in local feature e xtraction and GNN’ s capacity for relational reasoning to achie v e better performance in comple x image classication assignments such as detecting plant disease. The moti v ation for this study is also rooted in our earlier w ork on agri-graph neural netw ork (Agri- GNN), which le v eraged graph-structured learning to capture spatial and relational dependencies in agricultural prediction tasks, outperforming con v entional deep learning approaches for rice yield forecasting in Indonesia [27]-[30]. In this study , we collected primary data from 1,250 cocoa leaf photos collected in Jember Re genc y . Pre vious research related to cocoa disease diagnosis includes e xpert system–based approaches such as [31]; ho we v er , to the best of our understanding, there has been no study that specically applies machine learning methods to VSD disease in cocoa. Therefore, the research g ap, most research on cocoa leaf disease detection i s still limited to pure CNN approaches that only e xtract local spatial features without considering the relationships between leaf areas [32], [33]. Meanwhile, GNN-based models ha v e sho wn the ability to understand relationships between nodes, b ut are unable to e xtract comple x visual features in natural leaf images as f ar as we understand, there has been no research that combines these tw o approaches specically for detecting VSD disease in cocoa lea v es. Therefore, this study proposes a h ybrid CNN-GNN model that utilizes CNN as a local feature e xtractor and GNN as a relationship modeler between leaf areas to be used as a model for classifying the dataset we ha v e collected. This approach is e xpected to o v ercome the limitations of each model by producing a more comprehensi v e spatial-relational representation. 2. METHOD 2.1. Dataset and pr epr ocessing The 1,250 cocoa leaf images are di vided into tw o classes, namely Sehat (health y) and V ascular Str eak Diebac k (VSD) . The structure of the images is based on folders, enabling automatic label assignment with PyT orchs ImageFolder tool. All images were resized to a x ed resolution of 224 × 224 pix els to ensure uniform input di mensions. Images were then transformed into tensor representations by normalizing the pix el v alues to the range [0 , 1] . The dataset w as di vided into training v alidation and test sets at percentages of 70%, 10%, and 20% respecti v ely . respecti v ely ensuring balanced class distrib utions across all splits. A batch size of 32 w as used for mini- batch data loading. Data shuf ing w as enabled for the training set to enhance generalization performance and decrease sampling bias. Our research o w is sho wn in Figure 1. 2.2. CNN backbone ar chitectur e The core feature e xtractor w as a lightweight CNN. Three con v olutional blocks with progres si v ely deeper channels 32, 64, and 128 mak e up the architecture. Accordingly , the rectied linear unit (ReLU) ac- A gr aph neur al network fr ame work for vascular str eak diebac k r eco gnition (Slamin) Evaluation Warning : The document was created with Spire.PDF for Python.
196 ❒ ISSN: 2502-4752 ti v ation function comes after each con v olutional layer spatial do wnsampling using max pooling. An adapti v e a v erage pooling layer w as used at the end to create a x ed-length netw ork, feature representation independent of the spatial dimensions of the input. An output with 128 dimensions is the end result, each feature v ector acts as its visual embedding. The CNN backbone serv es tw o purposes , both as a feature encoder in the CNN–GNN model and as a standalone classier in the CNN-only baseline. Figure 1. Ov erall research o w illustrating the main stages of the study 2.3. Graph construction A graph structure w as dynamically b uilt to capture relational information between samples in e v ery mini-batch. P airwise cosine similarity w as calculated between e v ery sample batch of CNN feature embeddings. The top- K most similar samples were chosen as neighbors for each node, in this case study K = 5 is used. An undirected graph w as formed by creati ng edges in both directions, and the resulting structure w as represented using the edge-inde x format needed for graph neural netw orks. During training, relational reasoning between visually comparable samples is made possible by this batch-wise graph b uilding, which preserv es computing ef cienc y . 2.4. Graph neural netw ork module A tw o-layer graph con v olutional netw ork (GCN) w as used to create the graph-based reasoning com- ponent. A ReLU acti v ation function comes after the rst graph con v olution layer , which con v erts the 128- dimensional CNN features into a 256-dimensional hidden representation. The hidden representation is pro- jected into the output space that corresponds to the number of tar get classes by the second layer . The model may update feature representations based on int er -sample correlations and aggre g ate data from neighboring nodes by using graph con v olution processes. 2.5. T raining strategy The CNN-only and CNN-GNN models were trained for 20 epochs at a learning rate of 1 × 10 − 3 using the Adam optimizer . The training tar get w as the cross-entrop y loss function. Backpropag ation w as used to tune the model’ s parameters, and generalization performance w as e v aluated by tracking v alidation accurac y . Due to hardw are limitations, e v ery e xperiment w as carried out in a CPU-based setting. The suggested training conguration is still appropriate for assessing the ef fecti v eness of the suggested strate gy e v en in the absence of GPU acceleration. 2.6. Ev aluation metrics Se v eral metrics, including o v erall accurac y , precision, recall, and F1-score for each class, were used to e v aluate the model’ s performance. Additionally , class-wise prediction errors were analyzed using confusion matrices. The models’ ability to discriminate across v arious cate gorization thresholds w as e v aluated using recei v er operating characteristic (R OC) curv es and the associated area under the curv e (A UC). 3. RESUL TS AND DISCUSSION 3.1. Ov erall perf ormance of the CNN-GNN model The learning beha vior remained stable throughout the training process, with both training accurac y and v alidation accurac y sho wing meaningful impro v ements when observ ed across each epoch as presented in the training v ersus v alidation curv es. The maximum v alidat ion accurac y achie v ed by the model reached 95.2% as sho wn in Figure 2. This result can be used as a basis to state that the generalization capability is considered good, e v en though the training w as conducted in a CPU-based en vironment. It can be observ ed that the training accurac y and v alidation accurac y e xhibit relati v ely similar rising and f alling patterns. The results also sho w that the dif ference between the tw o v alues remains relati v ely small Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 194–204 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 197 for each epoch, indicating that no signicant o v ertting occurs. This suggests that the combination of con v olu- tional feature e xtraction and graph-based reasoning is able to learn meaningful representations from cocoa leaf images. Figure 2. T raining and v alidation accurac y curv es of the CNN–GNN model o v er 20 epochs 3.2. P er -class e v aluation and err or analysis Per -class metric e v aluation w as conducted to analyze the classication beha vior of the CNN-GNN model in greater detail. The e v aluation results sho w that the precision, recall, and F1-score v alues are relati v ely balanced for both classes. The complete results are presented as follo ws T able 1. These results indicate that the model is ef fecti v e in detecting infected lea v es while simultaneously maintaining a lo w f alse-positi v e error rate. T able 1. Classication performance metrics Class Precision Recall F1-score Support Sebat 0.94 0.95 0.95 64 VSD 0.95 0.93 0.94 61 Accurac y 0.94 125 Macro A vg 0.94 0.94 0.94 125 W eighted A vg 0.94 0.94 0.94 125 The confusion matrix is sho wn in Figure 3. More detailed information is pro vided by the confusion matrix related to classicat ion errors. From the confusion matrix, it can be stated that the classication is correct for most samples and only a small number of f alse ne g ati v es are observ ed. This conte xt is important in cocoa leaf disease detection, because if an infection occurs b ut is not detected, it may result in signicant agricultural production losses. 3.3. Discriminati v e capability analysis In Figure 4, which represents the R OC curv e of the CNN-GNN model, it is sho wn that approxima tely 0.94 is the A UC v alue of the model. Based on t his, the discriminati v e capability is considered suf ciently strong in performing classication of cocoa lea v es “Sehat” and those infected with VSD across v arious classication thresholds. The high A UC v alue indicates that the learned feature representations are ef fecti v e in separating the tw o classes, e v en when the decision threshold is adjusted. This further conrms the rob ustness of the proposed approach. A gr aph neur al network fr ame work for vascular str eak diebac k r eco gnition (Slamin) Evaluation Warning : The document was created with Spire.PDF for Python.
198 ❒ ISSN: 2502-4752 Figure 3. Confusion matrix of the CNN–GNN model on the v alidation dataset Figure 4. R OC curv e of the CNN-GNN model with an A UC v alue of approximately 0.94 3.4. Ablation study: CNN vs. CNN-GNN An ablation study w as conducted to e v aluate the contrib ution of the graph-based reasoni n g component. A comparison between the CNN-only model as a baseline and the CNN-GNN model is sho wn in Figure 5. In this section, it is interesting to observ e that the CNN-only model achie v ed a maximum v alidation accurac y of 97.6%, which is higher than the 95.2% maximum v alidation accurac y obtained by the CNN-GNN model. This result suggests that, for the dataset used, visual features alone are already highly discriminati v e, such that additional relational modeling through graph structures does not al w ays lead to performance impro v ement. Such results may be inuenced by se v eral f actors, including the possibilit y t h a t con v olut ional features are suf cient to separate the visual characteristics of Sehat cocoa lea v es and those infected with VSD. In addition, the ability of the GNN to model global relationships among samples may be constrained by the batch-wise graph construction strate gy . Interestingly , the C NN-only model achie v ed a slightly higher maximum v alidation accurac y (97.6%) compared to the CNN-GNN model (95.2%). These ndings indicate that, for the dataset used, visual features alone are highly discriminati v e, and thus additional relational modeling via graph structures does not necessar - Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 194–204 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 199 ily enhance performance. Another possible f actor is that the relati v ely moderate dataset size may reduce the ef fecti v eness of graph-based learning, as relational reasoning generally pro vides greater benets for lar ger or more ambiguous datasets. Ne v ertheless, these ndings do not diminish the rele v ance of the CNN-GNN approach. Instead, the results emphasize that graph-based reasoning is most benecial when inter -sample relationships pro vide com- plementary information be yond visual cues. Therefore, this ablation study of fers important insights into the conditions under which graph-enhanced models can operate optimally . Figure 5. Ablation study comparing the v alidation accurac y of the CNN-only and CNN-GNN models 4. CONCLUSION The study tests a h ybrid frame w ork using CNN and GNN to classify cocoa leaf diseases, focusing on distinguishing health y lea v es from VSD-af fected ones. A CNN feature embedding-based k-nearest neighbor graph represents sample connections in the proposed approach, which combines feature e xtraction with graph- based relational reasoning. The e xperiments sho w that the CNN-GNN model performs reliably with v alidation accurac y o v er 95%, balanced precision and recall across classes, and high R OC-A UC, CNN had slightly higher maximum v alidation accurac y than the baseline CNN model in an ablation study . This indicat es that visual features alone are highly ef fecti v e at distinguishing cate gories in the dataset, making graph-bas ed reasoning unnecessary for cons istent performance impro v ements. These ndings emphasize the importance of empirical assessment and ablation analysis in determining architectural element impact. This study illuminates graph- based deep learning model rele v ance in agricultural imagery . Graph-based reasoning uses sample relationships, b ut its ef fecti v eness depends on the dataset’ s characteristics, graph construction methods, and data size. W e will in v estig ate global graph constructi on methods, alternati v e similarity metrics, and lar ger , more di v erse datasets to better understand when CNN-GNN models outperform con v olutional methods. A CKNO WLEDGEMENT This re search w as supported by the Uni v ersitas Jember under an international collaboration schem e with Uni v ersiti Malaysia T erengg anu, Malaysia. The authors e xpress their gratitude to Uni v ersiti Malaysia T erengg anu for pro v i ding graph theory e xpertise that signicantly contrib uted to the success of this interna- tional research project. FUNDING INFORMA TION This research w as funded by the Uni v ersitas Jember under an international collaboration research grant scheme numbered 2956/UN25.3.1/L T/25 A gr aph neur al network fr ame work for vascular str eak diebac k r eco gnition (Slamin) Evaluation Warning : The document was created with Spire.PDF for Python.
200 ❒ ISSN: 2502-4752 A UTHOR CONTRIB UTIONS ST A TEMENT This journal uses the Cont rib utor Roles T axonomy (CRediT) to recognize indi vidual author contrib u- tions, reduce authorship disputes, and f acilitate collaboration. Name of A uthor C M So V a F o I R D O E V i Su P Fu Slamin ✓ ✓ ✓ ✓ ✓ Rizk y Alf anio Atmok o ✓ ✓ ✓ ✓ ✓ Antonius Cah ya Prihandok o ✓ ✓ ✓ ✓ ✓ Muhammad Ariful Furqon ✓ ✓ ✓ ✓ ✓ Qurrota A ’yuni Ar Ruhimat ✓ ✓ ✓ ✓ ✓ Annisa Fitri Maghiroh Harvyanti ✓ ✓ ✓ ✓ ✓ Bayu T aruna W idjaja Putra ✓ ✓ ✓ ✓ ✓ Roslan Hasni ✓ ✓ ✓ ✓ ✓ C : C onceptualization I : I n v estig ation V i : V i sualization M : M ethodology R : R esources Su : Su pervision So : So ftw are D : D ata Curation P : P roject administration V a : V a lidation O : Writing - O riginal Draft Fu : Fu nding acquisition F o : F o rmal analysis E : Writing - Re vie w & E diting CONFLICT OF INTEREST ST A TEMENT Each of the authors of this research declare no conict of interest. D A T A A V AILABILITY The data of this study has been collected directly by the authors in the cocoa plantations in Jember , Indonesia. The authors ha v e recorded directly the visual appearance of cocoa lea v es on a cocoa plant and replaced the background display with white by co v ering the background with perfect white paper and then sa ving it as image data of 1,250 cocoa leaf displays with white background image. REFERENCES [1] H. Hendra, A. W ibo w o, and E. Surya nti, “Fung al pathogens associated with v ascular streak dieback (VSD) disease on cacao in Special Re gion of Y ogyakarta Pro vince, ” AgroScience , v ol. 3, no. 2, pp. 60–67, 2019. [2] G. J. Samuels, A. Ismaiel, A. Rosmana, D. Guest, S. L. Daniel, V . M. Phan, and J. D. Rogers, “V ascular streak dieback of cacao in Southeast Asia and Melanesia: Identifying the pathogen, ” Mycologia , v ol. 104, no. 5, pp. 1086–1099, 2012, doi: 10.3852/11-289. [3] G. J. Samuels, “V ascular -Streak Dieback of cacao in Southeast Asia and Melanesia, ” Journal of Plant P athology , 2012. [4] Badan Pusat Statistik (BPS), Indonesian Cocoa Statistics 2023 , Jakarta, Indonesia, 2024. [Online]. A v ailable: https://www .bps.go.id/en/publication/2024/11/29/ed255af0c9059f288fb7e1de/indonesian-cocoa-statistics-2023.html [5] A. Rosmana, A. Nurmansyah, A. Suryadi, and D. A. Santoso, “Dynamic of V ascular Streak Dieback Diseas e Incidence on Cacao, ” Agronomy , v ol. 9, no. 10, p. 650, 2019, doi: 10.3390/agronomy9100650. [6] G. Rangel, J. C. Cue v as-T ello, J. Nunez-V arela, C. Puente, and A. G. Silv a-T rujillo, “ A surv e y on Con v olutional Neural Netw orks and their performance limitations in image recognition tasks, ” Journal of Sensors , v ol. 2024, 2024, doi: 10.1155/2024/ArticleID. [7] S. P . Mohanty , D. P . Hughes, and M. Salath ´ e, “Using deep learning for image-based plant disease detection, ” Frontiers in Plant Science , v ol. 7, p. 1419, 2016, doi: 10.3389/fpls.2016.01419. [8] E. C. T oo, L. Y ujian, S. Njuki, and L. Y ingchun, “ A comparati v e study of ne-tuning deep learning models for plant disease identication, ” Computers and Electronics in Agriculture , v ol. 161, pp. 272–279, 2019, doi: 10.1016/j.compag.2018.03.032. [9] K. S. K ouassi, M. Diarra, K. H. Edi, and B. J.-C. K oua, “Detection of cocoa leaf diseases using the CNN-based feature e xtractor and XGBoost classier , ” Open Journal of Applied Sciences , v ol. 14, pp. 2955–2972, 2024, doi: 10.4236/ojapps.2024.1410193. [10] W . T ang, “Re vie w of Image Classication Algorithms Based on Graph Con v olutional Netw orks, ” EAI Endorsed T ransactions on AI and Robotics , 2023, doi: 10.4108/airo.3462. [11] U. Nirosha and G. V ennil a, “Enhancing crop yield prediction for agriculture producti vity using federated learning inte grating with graph and recurrent neural netw orks model, ” Expert Systems with Applications , 2025, doi: 10.1016/j.esw a.2025.128312. [12] D. Sumathi, R. Karthik e yan, and S. Prabhu, “Early plant disease detection using graph isomorphic netw orks: Enhancing crop yield through leaf analysis, ” Journal of Computer Science , v ol. 21, no. 9, pp. 2065–2073, 2025, doi: 10.3844/jcssp.2025.2065.2073. [13] S. Maruthai, R. S. K umar , and P . Balasubramanian, “Hybrid vision graph neural netw orks-based early detection and protection for cof fee crop pests, ” Scientic Reports , 2025, doi: 10.1038/s41598-025-96523-4. [14] E. A. Mahareek, M. Al-Hamdan, and T . Alshammari, “Inte grating con v olutional, transformer , and graph neural netw orks for agri- cultural remote sensing applications, ” Agriculture , 2025, doi: 10.3390/agriculture1510353. Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 194–204 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 201 [15] M. H. Rahman, L. Chen, and Y . Zhang, “ A graph con v olutional netw ork approach for h yperspectral plant ph ysiological prediction, ” Computers and Electronics in Agriculture , 2025. [16] J. P . Rodrigues and J. L. Carbonera, “Graph Con v olutional Netw orks for Image Classication: Comparing Approaches for Building Graphs from Images, ” in Proc. SCITEPRESS , 2024. [17] T . N. Kipf and M. W elling, “Semi-Supervised Classication with Graph Con v olutional Netw orks, ” in Proc. Int. Conf. Learn. Rep- resentations (ICLR) , 2017. [Online]. A v ailable: https://arxi v .or g/abs/1609.02907 [18] W . Zhao, T . Jiang, and L. Zhang, “Structural graph learning for image classication, ” Neural Netw . , v ol. 139, pp. 179–190, 2021, doi: 10.1016/j.neunet.2021.02.021. [19] J. P ark, H. Kim, and S. Lee, “Graph attention netw orks for image classication based on visual relationship reasoning, ” IEEE Access , v ol. 11, pp. 87921–87933, 2023, doi: 10.1109/A CCESS.2023.3304529. [20] Y . W u, X. Li, and J. Chen, “Fine-grained image classication using adapti v e graph con v oluti onal netw orks, ” P attern Recognit. Lett. , v ol. 182, pp. 1–9, 2024, doi: 10.1016/j.patrec.2024.01.003. [21] Y . W ang, Y . Sun, Z. Liu, S. E. Sarma, M. M. Bronstein, and J. M. Solomon, “Dynamic graph CNN for learning on point clouds, ” A CM T rans. Graph. , v ol. 38, no. 5, pp. 1–12, 2019, doi: 10.1145/3326362. [22] X. Zhang, H. Xu, and Y . Zhao, “Hybrid CNN–GNN model for brain tumor classication using MRI images, ” IEEE Access , v ol. 10, pp. 98212–98223, 2022, doi: 10.1109/A CCESS.2022.3198214. [23] L. Chen, R. Zhao, and H. W ang, “Graph con v olutional and con v olutional neural netw ork h ybrid for plant disease detection, ” Comput. Electron. Agric. , v ol. 214, Art. no. 109015, 2024, doi: 10.1016/j.compag.2024.109015. [24] X. Xu, L. Ma, and Q. Zhao, “Hybrid CNN–GNN frame w ork for scene graph generation in visual understanding, ” IEEE T rans. Image Process. , v ol. 30, pp. 10367–10379, 2021, doi: 10.1109/TIP .2021.3128741. [25] Y . Li, J. Chen, and P . Xu, “CNN–GNN h ybrid netw ork for breast cancer histopathological image classi cation, ” P attern Recognit. , v ol. 138, Art. no. 109388, 2023, [26] Z. Fei, J. Guo, H. Gong, L. Y e, E. Attahi, a nd B. Huang, “ A GNN architecture with local and global-attention feature for image classication, ” IEEE Access , v ol. 11, pp. 110221–110233, 2023, doi: 10.1109/A CCESS.2023.3285246. [27] R. A. Atmok o, S. Slamin, A. C. Prihandok o, M. A. Furqon, Q. A. Ar Ruhimat, and B. T . W . Putra, “Indonesia rice yields forecasting using Agri-Graph neural net w ork, ” in Proc. 9th Int. Conf. Man-Machine Systems (ICoMMS) , 2025, doi: 10.1109/ICoMMS66553.2025.11200184. [28] A. Gupta and A. Singh, “ Agri-GNN: A no v el genotypic-topological graph neural netw ork frame w ork b uilt on GraphSA GE for optimized yield prediction, ” arXi v:2310.13037, 2023. [Online]. A v ailable: https://arxi v .or g/abs/2310.13037 [29] K. W ang et al., “Maize yield prediction with trait-missing data via bipartite graph neural netw ork, ” Front. Plant Sci. , v ol. 15, Art. no. 1433552, 2024, doi: 10.3389/fpls.2024.1433552. [30] F . Y ang et al., “Prediction of corn v ariety yield with attrib ute-missing data via graph neural netw ork, ” Comput. Electron. Agric. , 2023. [31] A. Franz, N. Nurlaila, and P . Y . Andayani, “Expert system for diagnosing cocoa diseases using the Demp- ster–Shafer method, ” Politeknik Pertanian Ne geri Samarinda , v ol. 1, no. 1, Mar . 2020. [Online]. A v ailabl e: htt ps://e- journal.politanisamarinda.ac.id/inde x.php/tepian/article/vie w/49/33 [32] K. Hu, G. Coleman, S. Zeng, Z. W ang, and M. W alsh, “Graph W eeds Net: A graph-based deep learning method for weed recogni- tion, ” Comput. Electron. Agric. , v ol. 174, Art. no. 105520, 2020, doi: 10.1016/j.compag.2020.105520. [33] R . A. Meshram and A. S. Alvi, “Design of an iterati v e method for crop disease analysis incorporating graph attention with spatial-tempora l learning and deep Q-netw orks, ” Int. J. Intell. Eng. Syst. , v ol. 17, no. 3, pp. 706–718, 2024, doi: 10.22266/ijies2024.0630.55. BIOGRAPHIES OF A UTHORS Pr of . Drs. Slamin, M.Comp.Sc., Ph.D . is a Professor of Computer Science at the Depart- ment of Informatics, F aculty of Computer Science, Uni v ersity of Jember , Indonesia. He recei v ed his Drs. de gree from the Uni v ersity of Jember in 1991, the M.Comp.Sc. de gree from the Uni v ers ity of Ne wcastle, Australia, in 1998, and the Ph.D. de gree from the same uni v ersity in 2002. His research interests include graph theory , graph labeling, combinatorics, cryptograph y , machine learning, net- w ork optimization, and computational intelligence. He has taught a wide range of courses, includ- ing discrete mathematics, graph theory , professional issues, decision support systems, and adv anced courses in informatics and information systems. He has been ac ti v ely in v olv ed in numerous national and international research grants, particularly in graph theory applications, c ybersecurity , cryptog- raph y , graph neural netw orks, smart f arming, optimization, and netw ork modeling. Throughout his career , he has published e xtensi v ely in reputable international journals such as SOIC, Opuscula Math- ematica, EJGT A, Symmetry , Heliyon, APJIS, IJ AERS, and Kraguje v ac Journal of Mathematics. He has also contrib uted to international conferences and collaborati v e research projects, including stud- ies on antimagic labeling, metric dimension, dominating sets, and netw ork optimization. He has authored books, including T eori Gr af dan Aplikasinya , and has secured multiple cop yrights related to graph theory and optimization systems. He has also serv ed as a speak er , re vie wer , and collaborator in v arious scientic forums. He can be contacted at email: slamin@unej.ac.id. A gr aph neur al network fr ame work for vascular str eak diebac k r eco gnition (Slamin) Evaluation Warning : The document was created with Spire.PDF for Python.
202 ❒ ISSN: 2502-4752 Pr of . Drs. Antonius Cah ya Prihandok o, M.A pp.Sc., Ph.D . is a Professor of Computer Science at the Department of Informatics, F aculty of Computer Science, Uni v ersity of Jember , In- donesia. He recei v ed the Drs. de gree from the Uni v ersity of Jember in 1992, the M.App.Sc. de gree from James Cook Uni v ersity , Australia, in 1999, and the Ph.D. de gree from the s ame uni v ersity in 2015. His academic e xpertise includes cryptograph y , graph theory , graph labeling, netw ork optimiza- tion, information security , and mathematical modeling. He teaches a broad range of under graduate and postgraduate courses, including Modern Cryptograph y , Information Systems Security Manage- ment, Ethics and Professional Issues, Research Methodology , Algebra, and Basic Mathematics. He has be en acti v ely in v olv ed in numerous national and international research grants focusing on an- timagic labeling, asymmetric cryptograph y , graph neural netw orks, smart f arming technologies, net- w ork topology optimization, and machine learning–based security systems. e has published widely in reputable journals such as IJECE, Indonesian Journal of Combinatorics, IJCNIS, Internati onal Jour - nal on Adv anced Science, Engineering and Informati on T echnology , Journal of Ph ysics: Conference Series, and se v eral mathematics and information systems journals . His scholarly contrib utions also include research on stream cipher construction, rainbo w antimagic coloring, DRM systems, learning models, and optimization for netw ork infrastructure. He has also serv ed as a principal in v estig a- tor and collaborator in v arious international research program s, particularly i n biometric encryption, transportation netw ork analysis, and c yber resilience. His academic output includes arti cles, con- ference proceedings, and a book chapter on group and ring theory . He can be contacted at email: antoniuscp.ilk om@unej.ac.id. Rizk y Alfanio Atmok o, S.Si., M.Sc. is a lecturer at the Department of Informatics, F aculty of Computer Science, Uni v ersity of Jember . He recei v ed the S.Si. de gree from Uni v ersitas Gadjah Mada in 2016 and the M.Sc. de gree from the same uni v ersity in 2019. His research interests include quantum computing, graph databases, kno wledge graphs, spatial data analysis, cryptogra- ph y , and computational modeling. He teaches se v eral under graduate courses such as line ar algebra, statistics, programming languages, theory of automata, graph databases, kno wledge graphs, modern cryptograph y , basic mathematics, geographic information systems, and introductory GIS. His aca- demic contrib utions also i nclude the de v elopment of graph-based learning module s and the inte gra- tion of quantum information conce pts into computational frame w orks. Rizk y has published research in national journals and conference proceedings, including w orks on quantum-state encryption, spa- tial tuberculosis-case mapping, and agricultural forecast ing using graph neural netw orks (GNN). His publication record includes contrib utions to the International J ournal of Multidisciplinary Sciences and Art s , J-TIT , REMik , and IEEE conferences. He has been acti v e in community service acti vi- ties, such as serving as the lead trainer for the Informatics National Science Olympiad preparation at SMAN 2 Jember (2024). He also holds a re gistered cop yright for the academic information system F asilk omP artner s . He can be contacted at email: izk yaatmok o@unej.ac.id. Muhammad ‘ Ariful Fur qon, S.Pd., M.K om. is an Assistant Professor at the Department of Informatics, F aculty of Computer Science, Uni v ersity of Jember , Indonesia. He earned his S.Pd. de gree from Uni v ersitas Ne geri Malang in 2016 and the M.K om. de gree from Instit ut T eknologi Sepuluh Nopember (ITS) in 2019. His research interests include articial intelligence, deep learning, g ame de v elopment, graph databases, kno wledge graphs, machine learning for agriculture, and intel- ligent systems. He teaches se v eral courses, including statistics, graph databases, kno wledge graphs, articial intelligence, g ame design and de v elopment, g ame intelligence, g ame engine design, and research methodology . His academic acti vities al so encompass the de v elopment of intelligent sys- tems for agriculture, computational modeling, and educational technology . He has been in v olv ed in numerous research projects, including Deep Kno wledge Graphs for rice pest and disease diagnosis, machine-learning–based food security predi ction, and optimization in agroindustry supply chains. He has serv ed as principal in v estig ator in se v eral grants such as K eRis–DiMas and Dosen Pemula programs. He acti v ely contrib utes to community service, focusing on digital literac y , SDGs village de v elopment, ICT -based citizen services, and the implementation of virtual laboratory media for impro ving higher -order thinking skills. His publications include articles in national journals, interna- tional journals, and proceedings on topics such as image se gmentation, data visualization, kno wledge graph modeling, intelli gent assistants, and ICT for agriculture. He also holds multiple intellectual property rights, including eXperT ani 1.0, Deep Kno wledge Graph for rice pests and diseases, and se v eral ICT -based educational tools. He can be contacted at email: ariful.furqon@unej.ac.id. Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 194–204 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 203 Qurr ota A ’yuni Ar Ruhimat, S.Pd., M.Sc. is a lecturer at the Department of Infor - matics, F aculty of Computer Science, Uni v ersity of Jember , Indonesia. She recei v ed her Bachelor of Education de gree from the Uni v ersity of Jember in 2013 and her Master of Science de gree from Uni v ersitas Gadjah Mada in 2016. Her academic and res earch interests include applied mathematics, optimization, data science, data visualization, operational research, and netw ork topology modeling. She teaches courses such as statistics, basic mathematics, data science, data visualization, operational research, and research methodology . Her research acti vities co v er netw ork optimization, genetic al- gorithms, smart tourism digitalization, website security , and mathematical optimization for public health and nutrition programs. She has been acti v ely in v olv ed in nationally funded research, includ- ing optical distrib ution point optimization using genetic algorithms, access point topology modeling for smart tourism, and simple x-based optimi zation for nutritional balance analysis. She has also contrib uted to community service programs focusing on mathematics education, social media ana- lytics for SMEs, web-based school systems, augmented reality learning media, and rural economic empo werment. Her scholarly w orks include publi cations in journals and conference proceedings on netw ork topology , OSPF-based IoT monitoring systems, and TSP-based optimization. She holds a re gistered c op yright for educational material in computational intelligence. She can be contacted at email: qurrotaaar@unej.ac.id. Annisa Fitri Maghr oh Har vyanti, S.Si., M.Si. is a lecturer at the Department of Infor - matics, F aculty of Computer Science, Uni v ersity of Jember , Indonesia. She obtained her Bachelor of Science de gree in Mathematics from Uni v ersitas Gadjah Mada in 2020 and her Master of Science de gree in Mathematics from Uni v ersitas Jember in 2024. Her academic interests include applied mathematics, mathematical modeling, machine learning, computer vi sion, and data science, with a particular emphasis on de v eloping optimization-dri v en machine learning models. She teaches se v eral courses i n informatics, including Image processing, computer vision, articial intelligence, model- ing and simulation, ba sic mathematics , statistics, and linear algebra. Her academic e xperience also includes serving as a lecturer at Politeknik Ne geri Jember , a research assistant in precision agricul- ture, and a teaching assistant in computational mathematics. Her research portfolio spans AI-based agricultural diagnostics, Y OLO-based disease detection, con v olutional neural netw orks for cocoa leaf disease recognition, and multilingual sentiment analysis using BER T . She has been in v olv ed in projects inte grating dominating set theory with deep learning models, bridging theoretical mathe- matics and computational intelligence. Be yond academics, she has w ork ed as a tutor and subject e xpert for se v eral national EdT ech platforms, including Zenius, Snapquiz, Qanda, and OLC Edukasi. She has also been acti v e in community service programs as a v olunteer educator and humanitar - ian w ork er . Her broader interests include science communication, digital learning, and data-dri v en inno v ation. She can be contacted at email: annisafmh@unej.ac.id. Pr of . Bayu T aru na W idjaja Putra, Ph.D . is a Full Professor at the Department of Agricultural Engineering, F aculty of Agricultural T echnology , Uni v ersity of Jember , Indonesia. He recei v ed the B.Agr .T ech de gree from the Uni v ersity of Jember in 2008, the M.Eng. de gree in Agri- cultural Systems and Engineering from the Asian Institute of T echnology (AIT), Thailand, in 2013, and the Ph.D. de gree in the same eld from AIT in 2017. His research interests include precision agriculture, agricultural machine vision, remote sens ing for agriculture and forestry , wireless sensor netw orks (WSN), IoT -based monitoring, sensor standardization, deep learning, and machine learn- ing. He has authored numerous high-impact publications in Q1–Q3 international journals, including Measur ement , Information Pr ocessing in Agricultur e , Pr ecision Agricultur e , Data in Brief , F or est Science and T ec hnolo gy , Micr opr ocessor s and Mi cr osystems , and J ournal of Biosyst ems Engineer - ing . His research outputs include pioneering w orks on lo w-cost sensi ng systems, IoT -based h ydro- ponics monitoring, deep-learning–based agricultural disease detection, optical sensing for plant nu- trients, and remote sensing algorithms for plantation management. He is also an acti v e in v entor with multiple patents and cop yright s in sensing systems, optical calibration apparatus, chloroph yll mea- surement technologies, and precision agriculture softw are platforms. His w ork Agriino Handheld Nutrient Sensing System and SP AD-CAM has been wide ly recognized nationally . He is the founder and in v entor at Pr ecision Agriculture Indonesia Ltd., and has held leadership roles including Head of ITC Serv e UNEJ, Coordinator of Scientic Publication at LP2M UNEJ, and Head of the Laboratory of Precision Agriculture and Geoinformat ics. He is a member of ISP A, IEEE Computational Intel- ligence Society , and PER TET A. His a w ards include the prestigious KWEF–Japan Best Researcher A w ard (2019). He can be contacted via email: bayu@unej.ac.id. A gr aph neur al network fr ame work for vascular str eak diebac k r eco gnition (Slamin) Evaluation Warning : The document was created with Spire.PDF for Python.