Detection of Blueberry Freshness and Maturity Stages Using Support Vector Machine with Hybrid Color-Texture Features

Inconsistent blueberry quality evaluation in an industrial environment is a challenge to food safety and waste management. This paper introduces a two-stage classification approach based on Support Vector Machines (SVM) for automated quality evaluation. The main aim is to guarantee consumer safety by accurately distinguishing between rotten fruits while also evaluating maturity levels. Our approach uses a tailored hybrid feature vector that incorporates six color features (HSV Mean and Standard Deviation) and five texture features (GrayLevel Co-occurrence Matrix-GLCM) derived from low-cost RGB images. The approach is designed as a two-stage SVM model: the Freshness classifier based on the complete 11-feature set, and the Maturity classifier optimized for color features. To counteract class imbalance, we used Aggressive Class Weighting (3.5:1), which proved effective in eliminating bias and improving classification accuracy. The approach showed a 97 % Recall value for the “Rotten” class, validating the effectiveness of this costoptimized feature design strategy for high-speed agricultural quality evaluation.