Plant Diseases and Insect Pests Recognition Algorithm Based on D-YOLOv3

To address the problem of existing deep learning-based object detection algorithms that are difficult to accurately identify small object detection of plant diseases and pests, an optimized D-YOLOv3 algorithm is proposed. This algorithm uses K-means clustering algorithm to cluster the plant disease and pest dataset and obtain suitable anchor boxes. The original YOLOv3 backbone network is optimized by using DenseNet’s dense connection method to achieve cross-layer transmission of image information. The experimental results show that the optimized D-YOLOv3 network achieves a mAP value of 83.12% in the test set detection, which is 3.15% higher than the original YOLOv3 network and 0.37% higher than the YOLOv4 network. The optimized D-YOLOv3 well preserves the shallow feature information, effectively reduces gradient disappearance, missed detection and false detection, and has better detection effect.