Dual Networks Based System for Detecting and Classifying Harmful Insects in Orchards

Halyomorpha Halys is a type of a dangerous insect for agricultural crops, in general, and especially for orchards. The detection and evaluation of their abundance are of great importance for starting the elimination procedure in ecological conditions. There are few studies conducted on insect detection using deep neural networks, but none are dedicated to the Pentatomidae family. The paper proposes an intelligent system for the detection of two similar species, Halyomorpha Halys and N ezara Viridula, the first being the dangerous one. As a novelty, the paper proposes an integrated intelligent system consisting of two interconnected neural networks trained on a custom data set using transfer learning. The first is configurated for insect detection (YOLOv.4-tiny) and the second is dedicated to insect classification (EfficientNet-B3). This approach has the goal of decreasing the values of false negative and false positive errors, thus increasing the detection performance.