An Integrated Deep Learning Framework for Soil Type Classification

Abstract

Accurate soil type classification is crucial for optimizing agricultural practices, aiding in crop selection, and guiding land management decisions. We evaluated the performance of various deep learning models, including CNN combined with EfficientNetB0, EfficientNetB4, ResNet50, MobileNetV2, and VGG19, for soil categorizing s using a dataset of 1269 samples. Our analysis reveals notable variations in model performance across different soil types, highlighting the importance of selecting appropriate models for specific classification tasks. MobileNetV2 demonstrates outstanding accuracy, achieving a remarkable 100% accuracy for Alluvial soil classification. Conversely, ResNet50 exhibits high precision and recall values of 0.94 and 0.97, respectively, for Black soil classification. Among the model combinations, CNN with ResNet50 and MobileNetV2 emerges as the most promising, achieving the highest overall accuracy of 90.00%. These findings underscore the significance of employing deep learning models for accurate soil classification which is a foundational element for both agriculture and environmental management. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.