Automatic porcine diarrhea viruses classification using pathological images and hybrid semantic neural network

Porcine epidemic diarrhea is a highly contagious intestinal disease in pigs, caused by various strains of porcine epidemic diarrhea virus (PEDV). The infection rate in suckling piglets can reach 100%. Manual analysis of pathological images is the primary method for distinguishing PEDV types. However, the diversity and complexity of PEDV lead to low efficiency and accuracy of detection. To address this issue, we propose a novel pathological image-based amplitude and phase hybrid semantic neural network (APHNet). APHNet introduces amplitude and phase concepts from quantum mechanical wave functions to represent the original features of the images and the semantic variations, respectively. It extracts distinct semantic information from tokens in images, while ensuring characteristic weight aggregation. We validated APHNet on a constructed piglet diarrhea database and demonstrated high performance, with an accuracy of 85.56%, recall of 85.07%, F1-score of 0.86, and an area under the receiver operating characteristic curve of 89.02%. These experimental results demonstrate that APHNet is effective for PEDV classification. Furthermore, visualization and expert comparison methods were used to analyze differences in the pathological phenotypes of various viruses. The results of this study provide valuable insights for improving animal health and automated production monitoring.