Identification and counting of cucumber downy mildew sporangia based on the improved YOLOV5

Cucumber downy mildew is caused by Pseudoperonospora cubensis, with a wide range, strong epidemic and serious incidence. The number of pathogenic sporangia is one of the main factors affecting the occurrence and spread of the disease. Therefore, timely detection of the number of sporangia is an important basis for formulating control strategies. However, complex sporangia images and small objects present great challenges to detection. To overcome these challenges, this paper proposes an improved You Only Look Once (YOLO)-V5 network for detection of Downy mildew sporangia. The specific implementation method are as follows: (1) The Cross Stage Partial Network (CSPNet) is replaced by GhostNet. (2) Modify the connection mode of Feature Pyramid Network (FPN) + Path Aggregation Network (PANet), and replace the feature map of the original information with a fine-grained feature map. (3) The Squeeze and Excitatiom (SE) module is added to the backbone to improve the detection ability of small target objects. On the spore dataset we constructed, the Average Precision (AP50, IoU $\in {\{0} $, 0.5}) result of this network is 87.24%, which is 15.50% higher than the original, the FPS is 125.6,which is 15.4 higher than the original. Therefore, the improved YOLOV5 network can effectively detect cucumber sporangia in time.