Lightweight improved yolov5 model for cucumber leaf disease and pest detection based on deep learning

Many diseases and pests affect cucumber leaves as they grow. A lack of prompt control may result in reduced yields or crop failure. Early and correct detection of cucumber leaf diseases and pests is essential to prevent cucumber yield reduction. Manual detection consumes a large amount of human resources and exhibits poor real-time performance. In addition, there is a lack of public datasets for cucumber leaf disease images in real-world scenarios. To this end, firstly, this study collected and established a new cucumber leaf pest and disease dataset in a real-world scenario. Secondly, to accurately detect cucumber leaf diseases and pests, an improved cucumber leaf disease and pest detection model was proposed from scratch based on the original YOLOv5l model. To reduce the model size, the Bottleneck CSP module was used instead of C3 as the backbone and neck network section. Owing to the reduction in the number of parameters, number of layers, and calculations, the detection impact was significantly improved, in addition to that, the improved model able to detect even a small leaf spot of disease. Finally, the authenticity of the proposed model is demonstrated by incorporating the convolutional block attention module (CBAM) into both the improved and original YOLOv5l model. The experimental results showed that the mean average precision (mAP) of the improved model was 80.10%; whereas, the precision and recall were 73.8%, 73.9%, respectively. The improved model weight occupies only 13.6 MB of the memory. The overall improved model performance outperformed that of the original YOLOv5 model.