احسان جهاني پور

In animal sciences and veterinary medicine, image processing has been used in many applications. Such applications can range from eva‎luating meat composition to identifying live animals, tracking, monitoring behavior, and measuring phenotypes (such as biochemical characteristics), like estimating body weight, body condition score, and gait. Most methods currently used to measure desired traits in animals require specialized personnel who need training to maintain measurement quality. Moreover, most measurement processes are time-consuming, stressful for animals, and costly for farmers, which can hinder aspects like animal welfare and economic reasons. Therefore, there is interest in developing automated methods for livestock monitoring and measuring desired traits, and many studies have been conducted on non-invasive methods using image processing technology. This study examines and compares the performance of YOLOv5, YOLOv6, YOLOv7, and YOLOv8 models in detecting cow behaviors. The main objective of this research is to eva‎luate the accuracy and efficiency of different YOLO network versions in detecting 11 classes of cow behaviors using the CVB dataset. The results show that the YOLOv8 model, with a recall of 96.9%, accuracy of 95.7%, and mAP@0.5 of 98.1%, has the best performance among the models studied, providing higher accuracy and detection rates in identifying various cow behaviors. These findings indicate significant advancements in new YOLO versions and their effective application in animal behavior analysis and detection

بينايي ماشين , يادگيري ماشيني , پردازش تصوير , كشف حيوانات , تشخيص , يولو

computer vision , machine learning , image processing , Detection , tracking animal , YOLO

Ehsan Jahani Por

Saeed Ebadollahi