基于GMA-YOLO11n的黄羽肉鸡行为多目标检测方法研究OA

Timely identification of abnormal conditions in broiler chickens through routine inspections is a crucial means to improve the efficiency of intensive poultry farming management.In intensive poultry farming,computer vision-based inspection technologies outperform traditional manual methods in both accuracy and efficiency,showing significant potential for development.However,in large-scale farms,challenges such as multi-scale targets and occlusion pose considerable difficulties for models.To address these issues,this study proposes an improved object detection method,termed GMA-YOLO11n(GSConv and multi-scale attention YOLO11n),based on the YOLO11n(you only look once)framework.Specifically,a GSConv lightweight convolution module is introduced into the Backbone to reduce computational complexity,and a high-resolution feature layer of 160×160 is added through multi-scale feature fusion to enhance the detection of small-scale and densely distributed targets.In addition,a squeeze-and-excitation(SE)channel attention module is incorporated before multi-scale feature inputs to strengthen the representation of key features.Experimental results demonstrate that the proposed model can effectively perform multi-class detection of daily behaviors,including drinking,feeding,and walking,as well as abnormal states of broiler chickens.The model achieves mean average precision values of 93.87%and 90.45%on Dataset I and Dataset Ⅱ,respectively,both outperforming the baseline model,while maintaining an inference speed that satisfies practical video inspection requirements.