The Research on Goji Berry Pest and Disease Image Recognition Based on Adversarial Generative Enhancement

To take on the challenge of identifying small samples of lycium barbarum pests and diseases, this study built a closed-loop system that combines CGAN-ADA and MobileNetV3-Small. High-quality samples were generated using an adaptive discriminator augmentation mechanism, with the FID value optimized to 51.2 and the training dataset expanded by 200%. A multi-scale feature fusion strategy was employed to enhance classification accuracy, achieving an accuracy rate of 86.4% on the test dataset. An interactive three-zone interface was developed to support batch identification, with response latency below 0.5 seconds and confidence levels consistently above 95%. The system demonstrated a significant improvement in recognition efficiency across six major pest categories, including thrips, providing a deployable technical solution for field pest control.