Insect Classification Using Spatial Pyramid Pooling and Convolution Neural Network

This study aims to enhance the accuracy and efficiency of insect classification using spatial pyramid pooling convolutional neural networks (SPP-CNN). Improving traditional convolutional neural networks has resulted in poor performance when dealing with the diversity of insects and images of various sizes. The dataset, provided by a collaborating company, includes six different insect categories. Given the relatively limited number of samples in the existing dataset, we employed various data augmentation techniques to enrich our training samples, such as sharpening, adding noise, resizing, adjusting brightness, and randomly rotating images. Experiments demonstrate that SPP-CNN shows superior performance in insect classification tasks compared to efficientNetV2. The proposed architecture has achieved an accuracy rate of $83.8 \%$, which is $3.9 \%$ higher than the $79.9 \%$ accuracy of EfficientNetV2 architectures. This improvement indicates that incorporating the spatial pyramid pooling layer significantly enhances the accuracy of insect image classification. Moreover, the adaptability of SPP-CNN to various image resolutions without the need for retraining significantly boosts its practicality for real-world applications.