A muzzle-based self-enhancing cow recognition method using YOLO and LLM

Abstract

In the modern era, using the technology of image processing, people are beingfacilitated in many sectors. Therefore, we have decided to do research on cowrecognition using the methods of image processing. Because cows are prettyessential for human beings, and people from the entire world are being facilitatedby cows. Because of that reason, many technologies have been invented till nowfor the cow. Here we have presented an approach to identify the cow. Like thefingerprint of a human, the cow muzzle is also unique. Due to this reason, wecan recognize the cow based on the muzzle. To recognize muzzles, we have usedYolo and the Structural Similarity Index (SSI). Image processing is a sector ofArtificial Intelligence and data is quite important to train models. Moreover, totrain a model for image processing, it is needed to manage a high-configurationcomputer. But there are many regions in the world where people hardly get suchpowerful machines. To mitigate this issue, we have introduced a unique methodwhere we do not need a large amount of data for the image processing task. Toeliminate the necessity of collecting a huge amount of data, we included a self-enhancing feature, which helps the model to predict accurately and also helpsto improve performance by increasing the amount of data. However, withouta large amount of data, it will be possible to predict appropriately because ofthe presence of self-enhancing features. This certain program helps the modelto improve its capability for delivering faster responses. LLM is responsible toadd the self-enhancing feature. On the other hand, to find out the most suitableYolo model for our research, we have trained data with three different versionsof Yolo: Yolo v7, v8, and v9. We evaluated the performances of these threeversions in terms of Precision, Recall, mAP50, and mAP50-95 and found Yolov9 to be the best among them. It excels with a precision of 0.391, a recall of0.916, a mAP50 of 0.559, and a mAP50-95 of 0.307.