Rice planthopper is one of the most important pests of rice, which mainly includes white back planthopper, brown planthopper and small brown planthopper. In order to realize the rapid and accurate identification of rice planthoppers and prevent the same insect from being repeatedly identified and classified, the object detection algorithm combining image redundancy elimination and CenterNet network was proposed. Firstly, the field insect collection device independently developed by the team was used to automatically obtain insect images and make a data set. The data set was divided into four classes which included white back planthopper, brown planthopper, small brown planthopper and non-rice-planthopper. Secondly, for the live images with high similarity obtained by the field insect collection device, CenterNet with image similarity detection, image subtraction, image thresholding and bilateral filtering image redundancy elimination algorithms were combined, and a deep feature fusion network (deep layer aggregation, DLA) was selected, which was used as the backbone network to extract the characteristics of insects. Compared with the classic machine learning and deep learning models used in rice planthopper detection in the past, it had obvious advantages. The experiment results showed that for the preprocessed test set, the algorithm can not only quickly process insect images, but also can successfully solve the problem of insect repeated detection. The mean average precision was 88.1%, and the detection rate was 42.9f/s. The research effectively completed the identification and classification of the three types of rice planthoppers, and showed good generalization ability for insects collected in different time periods, which can be used for intelligent early warning and forecasting of rice pest outbreaks in the later period.