Research and application of improved Faster R-CNN model for pest identification

Aiming at the problem of low recognition accuracy of traditional pest detection methods in complex environments, this paper introduces dense convolutional network (DenseNet) and scale-transfer module technology into the Faster R-CNN model to optimize the structure of the Faster R-CNN model and improve the accuracy of the Faster R-CNN model in identifying pests. First, DenseNet169 and scale-transfer module technology are used to extract the feature map of pest image, and then the target bounding boxes is generated through Region Proposal Networks (RPN) network to obtain accurate pest feature information, and then the feature information in the target bounding boxes is input into the full connection layer to identify the target type. This experiment uses 600 pictures for training. Experimental results show that the improved model improves the recognition accuracy of the Faster R-CNN model, and the recognition accuracy reaches 93.4%, providing a new idea and method for pest identification. The improved model is applied to the early warning cloud platform for pest identification and analysis to identify agricultural and forestry pests. The model has an accuracy rate of over 90% for the identification of large individual pests, and has achieved relatively good application effects.