Early detection model for lameness disease in small broilers using YOLOv5n and Convolution neural network

Lameness in broilers is a walking impairment due to which broiler feel pain in legs while walking. Early detection of lameness in broiler is important which save farmers from huge economic loss. Daily observation of broiler’s health and growth is conducted manually, which subject to human error and labour-intensive. Therefore, there is need for an automatic model to detect lameness of broilers at early stage and predict their regular health status. This study addresses these problems using deep learning and computer vision for early detection of lame broilers. In this study, a model is proposed and its performance is tested for detecting small lame and healthy broilers. The proposed model consisted of two stages: (1) YOLOV5n model for lame broiler detection; and (2) The convolutional neural network model for classifying lame and healthy broilers. A video dataset of broilers is collected for this purpose and used to train the proposed model. In addition, the model is compared with the different version of YOLO (V3, V4, and V8).The results demonstrate the mAP of 98.3% at 0.5 IoU and precision, recall of 97.5%, 98.1% respectively which are better than the existing models. Overall, the proposed model has the potential to detect and classify healthy and lame broilers at an early stage.