Deep Convolutional Sequential Network for Digital Soil Mapping and Crop Recommendations Based on Soil Data
Abstract
Digital Soil Mapping involves using digital techniques to create detailed maps of soil properties across landscapes, utilizing data from remote sensing, soil sampling, and environmental variables. It is essential for effective land management, precision agriculture, and environmental monitoring, providing insights into soil characteristics such as texture, moisture, and nutrient content. However, challenges faced by existing models include inconsistent data quality and availability, the spatial variability of soil properties, difficulties in integrating multisource data, temporal dynamics of soil characteristics, computational complexity, and issues with user accessibility, which can hinder effective implementation and adoption. Hence, to address the existing challenges, this research introduces the explainable selective group enhanced attention-distributed deep convolutional sequential network (X-SGEA-D2CSN) for digital soil mapping (DSM) and crop recommendation. The proposed approach effectively addresses the DSM issues by meticulously capturing the subtle soil features and environmental patterns while filtering out irrelevant noise, ensuring better crop recommendations. Specifically, the proposed model effectively captures the complex patterns and long-term temporal dependencies associated with the soil properties crucial in understanding the dynamic agricultural environments. The experimental results demonstrate that the X-SGEA-D2CSN model achieves better performance, reporting a high accuracy of 94.17%, precision of 95.24%, and recall of 93.1% for 90% training due to the multi-scale feature learning with attention-based refinement, and optimizing the selection of relevant features.