TriFeat: a Feature Fusion Framework for Automated Image Classification in Agriculture (Case Study on Harumanis Mango Maturity)

This study introduces TriFeat, a handcrafted feature fusion framework that combines Histogram of Oriented Gradients (HOG), Gabor filters, and Local Binary Patterns (LBP) into a single discriminative descriptor for image classification. TriFeat is evaluated using multiple traditional machine learning classifiers to determine the most effective combination for Harumanis mango maturity recognition. To contextualize performance, a CNN-based model is included only as a comparative baseline and not as part of the fusion process. Experimental results show that TriFeat with SVM achieves the highest accuracy of 98.3%, outperforming both handcraftedonly and CNN baseline models. Although developed for agriculture, the framework provides a lightweight and computationally efficient feature fusion approach suitable for deployment on other areas such as medical imaging, smart farming, and industrial inspection. This makes TriFeat a scalable option for real-time, IoT-enabled quality monitoring systems.