Deep learning for insect detection and classification: a review

Insect detection is pivotal in agriculture, ecological monitoring, and public health, directly influencing crop yield and quality, ecosystem balance, and public health safety. However, traditional detection methods are constrained by labor intensity and inefficiency, making it difficult to meet detection demands for instantaneity, accuracy, and productivity. With the introduction of deep learning (DL), particularly the application of DL-based object detection techniques, significant advancements have been made in the field of insect detection. By summarizing classic insect datasets, highlighting the application of DL-based insect detection in three major scenarios, and comparing the latest methods in the past three years, this paper demonstrates the significant effectiveness of DL in improving accuracy, reducing false detection rates, and enhancing detection efficiency. Furthermore, this paper identifies the current challenges, including the difficulty in detecting small insects, insufficient detection accuracy in complex backgrounds and diverse poses, the trade-off between real-time performance and accuracy, and imbalanced data distribution. In response to these challenges, this paper explores corresponding strategies and outlines future research directions, such as fusing technology, enhancing dataset richness, and optimizing model structures.