Cucumber Leaf Diseases Detection Based on An Improved Faster RCNN

Crop disease is one of the main agricultural disasters. Using object detection technology for precision spraying is of great significance to environmental protection. In order to achieve cucumber leaf diseases detection under the natural environment, this paper proposes a cucumber leaf diseases detection method based on the improved faster region with convolutional neural network features (Faster RCNN). The model uses the residual convolutional neural network ResNet 50 as backbone to extract features from the input image, adopts feature pyramid networks (FPN) for feature fusion, then generates a series of proposal boxes through region proposal network (RPN), and ultimately generates bounding boxes and corresponding classes by the classification network. The results demonstrate that the network can reach a mAP value of 83
^%
, and the processing speed is 0.56 seconds per image. When the loss function is modified into GIOU, the mAP value boosts by 2%; when the anchor ratios are changed to 0.665, 1 and 1.53, mAP rises by 3.2%. The improved model has respectable generalization and robustness, which could be applied to the cucumber leaf diseases detection work well, gifting significantly promoting impact on farmland ecological protection.