An integrated intelligent system for early disease detection in swine using fused visual, acoustic, and infrared thermal features

Pig rearing is a key industry in global agriculture but is under severe threat from diseases like erysipelas, dermatitis, and swinepox. Traditional disease diagnosis is subjective, time-consuming, and in most cases cannot make an early diagnosis, leading to late treatment and high mortality. This study develops an intelligent system integrating visual, acoustic, and thermal data for early swine disease detection, enhancing accuracy and reliability. YOLOv11 detects healthy and sick pigs, CNN segments thermal features, U-Net classifies diseases, and MFCC-LSTM processes acoustic signals like coughing and screaming. Unlike unimodal methods that are sensitive to noise, occlusion, or weak early symptoms, the proposed multimodal fusion integrates complementary visual, thermal, and acoustic cues, enabling earlier and more reliable disease detection when individual modalities alone are insufficient. Visual analysis assesses health status, thermal infrared identifies abnormal temperatures, and audio detects respiratory issues. YOLOv11 achieved 99.60% accuracy, CNN thermal segmentation reached 98.92% precision and 95.83% recall, U-Net disease classification attained 99.61% accuracy, and MFCC-LSTM precision and recall were 94.82% and 94.14%. Combined, the system achieved 99.61% accuracy, recall, and F1-score, demonstrating robust early disease detection under varied farm conditions. The proposed system is capable of providing real-time, accurate detection of disease, enhancing swine health monitoring, animal welfare, and farm productivity. The integration of these high-performing models ensures timely detection of health anomalies, reducing mortality rates and optimizing herd productivity, thereby improving the economic and operational efficiency of livestock management.