Germination of chili pepper seeds is critical for crop yield and resource utilization. A high germination rate increases yield and effectively reduces resource wastage. This study collected 450 macroscopic images of chili pepper seeds and constructed a dataset for deep learning training through standardized germination experiments. Six deep learning models were evaluated to improve the chili pepper seed classification accuracy and germination rate. After comparing the performance of the models, MobileNet_v2 performed the best, not only having the fewest number of parameters but also achieving a 98.89% accuracy and 97.82% F1 score. The model improved the original germination rate from 87.33% to 100% on the test set, significantly optimizing the seed selection process

Crop diseases seriously affect food security, and traditional identification methods are inefficient and inaccurate. This paper proposes a GoogLeNet model with an attention mechanism. By integrating an attention module inside the Inception module, the recognition ability of subtle disease features and complex backgrounds is improved. Based on strict data preprocessing and enhancement, the proposed method achieves 87.75% accuracy on the AI Challenger 2018 crop disease dataset, which is better than the existing advanced methods, which verifies the effectiveness and practicability of the method and provides technical support for smart agriculture.

Animal pattern recognition from nighttime grayscale images is crucial for wildlife protection and ecological monitoring. Most of the current models suffer from a large parameter scale, making them unsuitable for deployment in resource-constrained environments. To address this challenge, this study proposes a multi-layer knowledge distillation approach based on the teacher Convnext model to improve the lightweight student model's classification performance effectively. Experimental results show that the parameters of the distilled CifarResNet20 model are only 0.27M, and the accuracy is 88.76%, which is superior to the traditional single-layer distillation and another tiny student model. The study confirms the efficiency and practical value of the proposed method in practical applications such as ecological monitoring.