An Classification Model of Helicoverpa assulta and Helicoverpa armigera Combining Spatial Transformation Network and Deep Convolutional Neural Network

Abstract

In order to effectively classify and identify the major pests such as Helicoverpa assulta and Helicoverpa armigera in tobacco, as well as to monitor and control them subsequently, a classification and recognition algorithm called Spatial Transformer Network combined with Deep Convolutional Neural Network (STN-DCNN) model is proposed for analyzing the images of these pests. Firstly, the images of two types of pests are preprocessed using STN to extract relevant features and obtain two different sub-images. These sub-images are then fed into a DCNN to obtain predictions for the images. Finally, the cross-entropy loss function is used to measure the difference between the prediction results and the corresponding labels. The results show that the model achieves effective classification and identification of the Helicoverpa assulta and Helicoverpa armigera with a recognition rate of 89.8%. This approach can provide valuable technical support for more accurate pest identification.