AUTHOR=Dai Fen , Wang Fengcheng , Yang Dongzi , Lin Shaoming , Chen Xin , Lan Yubin , Deng Xiaoling 

TITLE=Detection Method of Citrus Psyllids With Field High-Definition Camera Based on Improved Cascade Region-Based Convolution Neural Networks

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 12 - 2021

YEAR=2022

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2021.816272

DOI=10.3389/fpls.2021.816272

ISSN=1664-462X

ABSTRACT=Citrus psyllids is the only insect vector of citrus Huanglongbing(HLB), which is the most destructive disease in the citrus industry, so detecting and killing citrus psyllids is the key measure for citrus planting. To detect small target pest using field high-definition camera, a comprehensive machine vision solution is introduced in this paper. A sample enhancement method was proposed using semantic segmentation-based small target number enhancement methods and offline resampling methods. The Cascade R-CNN network was adopted for identification of citrus psyllids and was improved by using multi-scale training, combining CBAM attention mechanism with high-resolution feature retention network HRNet as feature extraction network, adding sawtooth ASPP structure to fully extract high-resolution features from different scales, and adding FPN structure for feature fusion at different scales. To mine difficult samples more deeply, online hard sample mining strategy was used in the process of model sampling. After a series of improvement measures, the improved Cascade R-CNN algorithm after training has an average recognition accuracy of 90.21% for Citrus psyllids. Compared with VGG16, ResNet50 and other common networks, the improved small target recognition algorithm obtains the highest recognition performance, which makes it possible and feasible to detect small target pests with field high-definition camera.