Effective Insect Recognition Based on Deep Neural Network Models in Complex Background

In biology, insects have the most species, number, distribution, and adaptability. Insect recognition is the basis of insect research and pest control. However, current insect recognition work mainly relies on few insect taxonomy experts. With the rapid development of computer technology, we can employ the computer instead of experts to distinguish insects accurately. To recognize insects effectively, especially the subtle differences between subcategories, we combined FGVC (Fine-Grained Visual Categorization) with deep learning, and applied Inception V3, VGG16_bn, and ResNet50 in the research of insect recognition and classification. In this paper, the experimental results showed that all of the three methods had high accuracy, the Inception V3 reached 98.69%, the VGG16_bn reached 97.80% and ResNet50 reached 97.94%. We also used label smoothing technology to reduce the errors caused by label errors and improve the accuracy of different models.