NEXT GEN Soybean Disease Detection Via CST-YOLO And Enhanced Yolov7– Transformer Fusion

Abstract

Soybean disease detection is necessary to keep crops healthy and improving productivity of agriculture. This paper proposes CST-YOLO, an innovative framework that integrates the YOLOv7 object detection model with a CNN–Swin Transformer architecture to boost preciseness and resilience of disease recognition. The framework incorporates customized improvements to YOLOv7, tailored specifically for soybean leaf analysis, while the Swin Transformer component captures complex spatial dependencies and strengthens the model’s ability to differentiate between plants that are healthy and those that are sick. Extensive experiments conducted on benchmark datasets validate the effectiveness of CST-YOLO, achieving superior detection performance compared to widely used ensemble- based techniques that are often considered state-of-the-art. The results highlight CST-YOLO as a reliable, efficient, and scalable solution for automated soybean disease diagnosis, with potential applications extending to the detection of leaf diseases in other crop species.