Target Detection And State Analysis Of Broilers Based On Deep Learning

With the development of social economy and the improvement of people’s living standards,consumers’ demand for meat products has also increased.As one of the essential meat products in human life,chicken accounts for 66% of poultry meat consumption in China.In the process of broiler breeding,the henhouse administrator needs to regularly check the henhouse and deal with sick chickens and dead chickens in time,but the efficiency is low and there are problems such as missed detection and false detection.In this study,broiler target detection under complex background was taken as the research target.Aiming at the problems of low intelligence of chicken house management and complex environment of chicken house in China,the research on broiler target detection and state analysis based on deep learning was carried out,and the network structure of deep learning model was improved.By comparison,the m AP value of the improved model increased from 91.43% to 96.41%,the detection accuracy of normal broilers was 96.53%,the recall rate was 96.32%,the detection accuracy of abnormal broilers was 96.28%,the recall rate was 95.86%,and the missed detection rate of broilers decreased from 9.23% to 4.81%.A mobile phone APP for broiler recognition based on Android henhouse was designed.The APP obtained the image information of broilers through the camera installed in the henhouse,counted the number of broilers,identified and recorded the abnormal broilers,and visualized them in the mobile phone APP.At the same time,the detection test was carried out in the henhouse,and the recognition accuracy was 94.28%.The main contents of this paper are as follows:(1)Construction of broiler dataset.Through field shooting and online collection of broiler related pictures,Open CV is used to enhance and expand the data set to improve the generalization ability of the model.Finally,the broiler image is manually labeled for the training of the deep learning model.(2)Selection of deep learning algorithms.Firstly,by comparing the four mainstream deep learning models of Faster R-CNN,SSD,YOLOv4 and YOLOv5,the data set is input into the four models for training,and then the training results are compared.Finally,YOLOv5 with better recognition effect is selected as the network model of this study.(3)The improvement of deep learning algorithm in complex background.Firstly,the attention mechanism is added to the YOLOv5 model to improve the acquisition ability of broiler features;secondly,multi-scale feature fusion technology is used to improve the detection ability of small target broilers and reduce the missed detection rate.Finally,the non-maximum suppression algorithm in the original model is modified to improve the detection effect of overlapping broilers.After testing,the model improved the recognition effect of broilers under complex background.(4)Development of mobile phone APP based on Android broiler identification.Through Android Studio platform and Java language,the design and development of broiler recognition APP is realized.The trained YOLOv5 model is transplanted into the mobile phone APP through the NCNN model transformation framework to realize the functions of broiler number statistics and abnormal broiler recognition.