Chicken Manure Disease Recognition Model Based on Improved ResNeXt50

Abstract

The feces of sick chickens often show a certain degree of morbidity. It is an effective method to identify sick chickens by deep learning. Concerning the environmental factors of chicken manure shooting that have a great influence, this paper proposes the introduction of a mixed attention mechanism into the ResNeXt50 network model and applies it to the chicken manure data set. While using residual structure learning, the model enhances the perception and learning ability of the input image. The results showed that the overall accuracy of the ResNeXt50-3A model on the test set reached 97.4%, which was 17.3, 1.5, 28.6, and 1.5 percentage points higher than the RegNetY_400MF, Vgg19, ViT (Vision Transformer) and ResNeXt50 models respectively. The ResNeXt50-3A model has high recognition accuracy and can provide support for the accurate identification of diseased chickens in all aspects of chicken breeding, replacing manual detection of diseased chickens.