A Late Fusion CNN Integrating RGB and Texture Features for Banana Ripeness Classification
Year: 2026
Type: conference-paper
Abstract: Banana ripeness classification plays an important role in automated quality assessment and supply chain management. However, most existing approaches rely solely on color information or handcrafted texture descriptors, which are often sensitive to illumination changes and surface variations. To address this limitation, we propose a Late Fusion Convolutional Neural Network (CNN) that integrates complementary RGB and texture features for robust ripeness recognition. The RGB stream employs a pretrained ResNet-50 backbone to capture semantic and color information, while the texture stream is a lightweight CNN trained on grayscale representations to extract fine-grained surface patterns. The two feature vectors are concatenated at the late fusion stage and processed by fully connected layers for final classification. Experiments conducted on a banana ripeness dataset demonstrate that the proposed model achieves an accuracy of 98.22%, outperforming traditional machine learning and single-stream CNN models. These results confirm that combining color and texture cues through a dual-stream fusion architecture significantly enhances classification accuracy and robustness under varying lighting and texture conditions.
Institutions: FPT University, Tay Do University
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: hybrid
OpenAlex ID: W7203667730