Prediction of Fruit Texture Features Using Deep Learning Techniques

• Published in: 2023 4th International Conference on Smart Electronics and Communication (ICOSEC) pp. 762 - 768
• Main Authors: Sangeetha, M., Kannan, S.Rajes, Boopathi, Sampath, Ramya, J, Ishrat, Mohammad, Sabarinathan, G.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.09.2023
• Subjects: Accuracy; Cameras; Deep learning; Forestry; GLCM; Image processing; K-Nearest Neighbor; Kappa; Machine learning algorithms; Prediction algorithms; Prototypes; Random Forest; RGB and Thermal Texture
• DOI: 10.1109/ICOSEC58147.2023.10276278

In this research, a prototype has been created to automatically collect the photos of fruits using a smartphone camera and a thermal camera, generating two distinct datasets: thermal and RGB. Fruit classification and bruise detection have both been accomplished using image processing, machine learning, and deep learning. To assess quality and shelf-life, a second set of RGB and thermal picture datasets was created. The suggested study enhances fruit categorization and flaw detection systems by utilizing multiple Grey level co-occurrence matrix (GLCM)-based texture characteristics. Using color-texture characteristics, pictures are classified using RF and KNN algorithms. RGB features were classified with 97.24 percent accuracy by K-Nearest Neighbor (KNN) and Random Forest (RF) classifiers, while GLCM-based features were classified with 96.40 percent accuracy by the RF classifier. Using thermal pictures, the KNN and RF algorithms' prediction accuracy was 81.70 percent and 84.12 percent, respectively. KNN and RF algorithms have kappa (percentage) values of 79.97 and 81.86, respectively. The random forest (RF) technique is hence the most appropriate for classifying fruit using a thermal picture dataset.