<![CDATA[Stunting is a chronic nutritional issue in children that can have long-term impacts on physical growth and cognitive development. Early detection is therefore important to support timely intervention. This study develops an image-based stunting detection approach that integrates HSV color descriptors, Gray Level Co-occurrence Matrix (GLCM) texture descriptors, and CNN features extracted with MobileNetV2. Images are preprocessed through resizing and Region of Interest (ROI) cropping based on bounding-box annotations. The handcrafted HSV and GLCM features are fused with CNN features at the feature level through vector concatenation before classification into stunting and non-stunting categories. This design was selected to preserve complementary low-level color-texture information and high-level semantic representations in a single classifier input. The hybrid model achieved a test accuracy of 84.39% with a stunting recall of 83%. Although the results indicate that multimodal visual descriptors can improve classification performance compared with single-feature approaches, the model still shows mild overfitting and was evaluated on a relatively limited dataset. In addition, inference efficiency and robustness to variations in imaging conditions were not yet quantitatively measured. Therefore, the present system should be interpreted as a proof of concept for objective, early, image-based stunting screening by healthcare personnel.]]>
Copyrights © 2026
