A Hybrid Deep–Handcrafted Feature Fusion Framework for Soil Image Classification and Intelligent Crop Recommendation

Soil is prime natural resources that affects ecosystem stability, environmental sustainability, and agricultural productivity. Precision agriculture and intelligent crop management depend heavily on accurate soil classification. Numerous studies have been proposed by various researchers to determine crop recommendations and soil classification. However, the fine-grained texture and color variations inherent in soil images make classification challenges. This study proposed a hybrid deep-handcrafted feature fusion framework that combines handcrafted descriptors like Local Binary Pattern (LBP) and Color Histogram with deep features based on Convolutional Neural Networks (CNNs). Initially, we applied a conventional classifier like Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Logistic Regression (LR), Random Forest (RF), and AdaBoost. Where RF achieved the best accuracy of 84.66%. In order to enhance the accuracy applied several pretrained transfer learning models, such as VGG16, ResNet50, MobileNetV2, InceptionV3, and DenseNet121.Based on performance, ResNet50 provides better accuracy 77.28% than other transfer learning models. The proposed hybrid fusion model demonstrated the superior performance, with 99.00% accuracy, whereas the CNN baseline model we developed and achieved 94.10% accuracy. Robustness was evaluated using precision, recall, F1-Score, and AUC metrics. In addition, a Graphical User Interface (GUI) was developed for real-time soil classification and crop recommendation enabling data-driven, sustainable agricultural practices.