Insects are one of the significant influences on global agricultural productivity. This study addresses the challenges of class imbalance and uneven data distribution in the IP102 dataset by investigating the integration of advanced data augmentation, multiple optimization algorithms, and transfer learning, employing EfficientNetV2 models for the insect classification task. The online augmentation strategies are integrated to reduce overfitting, multiple optimization algorithms are explored to improve model accuracy and leveraging knowledge of transfer learning to speed up training and reduce the need for expensive hardware. The optimization techniques’ performance varies significantly across all EfficientNetV2 models, according to experimental results. Visualization approaches are also used in the study to improve the interpretation of the image classification model’s output. Utilizing adamw optimization algorithms, EfficientNetV2 indicates the higher performance comparing to ResNet50, VGG16 and EfficientNetB6 for classification task in IP102 dataset.