DSCformer: Lightweight model for predicting soil nitrogen content using VNIR-SWIR spectroscopy

Chenxiao Li, Lifen Song, Lihua Zheng, Ronghua Ji

Abstract: Highlights•Innovative Lightweight Model: This paper introduces DSCformer, a novel lightweight model based on a new architecture that combines deep learning with spectral techniques.•Superior Performance: DSCformer outperforms other deep learning methods, including CNN, ResNet, VIT, MobileNet, and ShuffleNet.•Lightweight and Efficient: DSCformer is exceptionally lightweight with the fewest parameters and minimal computational requirements, making it highly efficient. It demonstrates fast CPU and GPU speeds.•Practical Application Potential: DSCformer has the potential to facilitate the practical deployment of deep learning models in soil spectral prediction.