Application of Deep Convolutional Neural Networks for Nutrient Mapping in Tobacco Rotation Systems through Image Analysis

Accurate detection and efficient management of soil nutrients in tobacco rotation systems are essential to improve tobacco yield and quality. Based on the existing deep convolutional neural networks of DenseNet, this work proposes a complex model that integrates multi-scale feature extraction, attention mechanism and residual enhancement module to perform multi-channel feature learning and cross-correction on field images. The model first extracts high-resolution and low-resolution features through multi-layer convolution and pooling structure, then uses the attention mechanism to weighted aggregate the features, and finally avoids gradient disappearance and strengthens the learning of key information through residual joining, so as to achieve accurate mapping and prediction of the spatial distribution of soil nutrients in tobacco rotation system. Experimental results show that the proposed model achieves higher detection accuracy and prediction stability in tobacco field nutrient mapping tasks, which provides a more feasible and accurate reference for scientific fertilization.