Adaptive feature fusion pyramid network for multi-classes agriculturalpest detection

ABSTRACT

Accurate and robust crop pest detection system is an important step in the reliable prediction of agricultural pests in precision farming communities, and has received high attention in many countries.In order to realize the automatic identification and detection of agricultural pests, the previous methods adopted the method based on image processing, resulting in low efficiency.Then, using hand-crafted feature descriptors, a machine vision-based crop pest detection method is introduced, which improves detection accuracy and detection speed.However, the manual function does not allow precise identification.Considering the powerful capability of Convolutional Neural Network (CNN) feature extraction, we developed a CNN-based method for multi-class pest detection in complex scenarios.In this paper, an adaptive feature fusion is introduced into the feature pyramid network to extract richer pest features.Then, an adaptive augmentation module is developed to reduce the information loss of the highest-level feature maps.Finally, a two-stage region-based convolutional neural network (R-CNN) is built to refine the predicted bounding box, and the pest category and location can be obtained for each image.We conduct extensive comparative experiments on the AgriPest21 dataset.Our method can achieve 77.0% accuracy, significantly outperforming other state-of-the-art methods, including SSD, RetinaNet, FPN, Dynamic R-CNN, Cascade R-CNN.

1. Introduction      

Agriculture is a basic industry that plays an important role in my country's economic development.The negative impact of frequent agricultural disasters is becoming more and more obvious.Crop pests are one of the main reasons for the decline in the quality and yield of agricultural products.Therefore, the effective control of agricultural pests has attracted extensive attention from academia and industry.In recent years, chemical pesticides have been widely used and become the most important means of controlling crop pests.However, farmers cannot identify pests,