Large-Scale Insect Pest Image Classification

— One of the main issues with agricultural production is insect attack, which leads to poor crop quality. Farmers, however, have a complicated and time-consuming task in detecting and categorizing insects. Therefore, research on an effective system for image-based automated insect classification is crucial. The conventional “softmax” function is utilized to determine the category for new image occurrences and minimize “cross-entropy” loss in the bulk of current research, which focuses on employing deep convolutional neural networks to categorize insect images. This paper presents a novel method for large-scale insect pest image classification by combining fine-tuning EfficientNets and Power Mean Support Vector Machine (SVM). First, EfficientNet models are fine-tuned and re-trained on new insect pest image datasets. The retrieved features from EfficientNet models are then utilized to create a machine learning classifier. In the network’s classification stage, the traditional “softmax” function is substituted with a non-linear classifier, Power Mean SVM. As a result, rather than “cross-entropy loss,” the training process focuses on reducing “margin-based loss.” Several benchmark insect image datasets were used to evaluate our proposed method. According to the numerical results, our method outperforms other cutting-edge methods for large-scale insect pest image categorization. With fine-tuning EfficientNets and Power Mean SVM, the classification accuracy of the proposed method for the Xie24, D0, and IP102 large insect pest datasets is 99%, 99%, and 72.31%, respectively. To our knowledge, these are the best performing image classification results for these datasets.