Accurate cotton diseases and pests detection in complex background based on an improved YOLOX model

Cotton diseases and pests are essential factors affecting the quality and yield of cotton in agricultural production. To meet the needs of intelligent agriculture development and solve the problems of low efficiency and poor reliability of cotton diseases and pests detection, this study proposed a real-time high-performance detection model based on improved YOLOX. The model introduced Efficient Channel Attention (ECA), hard-Swish activation function, and Focal Loss function into YOLOX, which improved the ability of the model to extract image features, solved the problem of sample imbalance, improved the detection speed and accuracy, and enhanced the detection effect of cotton diseases and pests. A total of 5760 manually labeled cotton diseases and pests images (including five kinds of red leaf blight, verticillium wilt, cotton spider mite damage, double-spotted leaf beetle damage, and brown spot disease) were used to fine-tune and test the model. The mean Average Precision (mAP) of cotton diseases and pests detection reached 94.60%, the precision was 94.04%, the F1-score was 0.90, and the FPS was 74.21. Furthermore, the results were compared with five classical object detection algorithms (Faster R-CNN, SSD, YOLOv3, YOLOv4, and YOLOv5). The comparative results showed that the mAP of the improved model was 11.50%, 21.17%, 9.34%, 10.22%, and 8.33% higher than the other five algorithms, and the detection speed can meet the real-time requirements. Finally, a cotton diseases and pests detection software was designed and developed based on the improved model, deployed on the smartphone to complete the real-time detection of cotton diseases and pests in the field environment. The improved model can effectively detect the infected area of cotton leaves in the field and provide theoretical reference and technical support for controlling cotton diseases and pests.