Affiliations:
1College of Artificial Intelligence, Zhongkai University of Agriculture and Engineering, Guangzhou, China
2Network and Information Technology Office, South China University of Technology, Guangzhou, China
Early detection of garlic diseases is essential for improving agricultural quality and productivity. This study presents a novel garlic disease identification model based on near-infrared (NIR) spectroscopy and a convolutional neural network, named ST-1DResNet (One-dimensional Residual Networks with Squeeze-and-Excitation and tanh activation). The model overcomes the vanishing gradient problem, adaptively adjusts channel weights, and efficiently extracts spectral features without requiring preprocessing or manual feature extraction. Experimental results show that ST-1DResNet achieves a classification accuracy of 97.75%, outperforming the original ResNet and four classical deep learning models by an average of 6.40%. Compared with traditional machine learning methods and optimized SVM models, it improves accuracy by 11.63% and 2.67%, respectively. The model is compact, computationally efficient, and supports fast training, making it suitable for deployment in resource-limited environments. Its strong generalization performance, validated using an external mango dataset, highlights its scalability. Overall, ST-1DResNet provides a practical, accurate, and non-destructive approach for crop disease detection, contributing to quality control and intelligent diagnosis in modern agriculture.
Garlic disease, Near-infrared spectroscopy, Convolutional neural network, Deep learning, Agricultural diagnosis
https://doi.org/10.21833/ijaas.2025.11.007
Zhang, R., Ye, T., Li, Z., & Deng, T. (2025). A garlic disease identification model based on near-infrared spectroscopy and an optimized ResNet. International Journal of Advanced and Applied Sciences, 12(11), 57–71. https://doi.org/10.21833/ijaas.2025.11.007