An Optimized Hybrid Deep Learning Approach for Accurate Fruit Image Classification

Le résumé fourni par la source

The field of fruit classification in computer and machine vision is growing rapidly. However, numerous deep learning approaches have been introduced for image classification, but they often encounter challenges that must be addressed. The effectiveness of the classification system relies on several factors and the selection of relevant features. In this paper, we propose an innovative hybrid deep learning framework integrating Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Gated Recurrent Unit (GRU) to classify fruit images accurately. The extraction and selection of optimal features are the key to achieving high classification accuracy. To achieve this, we leverage the power of CNN, RNN, and GRU, which are jointly employed in the automatic fruit classification process. The CNN was applied to extract spatial features from images. Then, an RNN was utilized to identify the most discriminative features, and finally, GRU performed final classification using the refined feature set from both CNN and RNN. Moreover, hyperparameters of the proposed model are optimized using TLBO-MCET. Empirical evaluations highlight the proposed method’s superiority over traditional methods such as Support Vector Machine (SVM), Feed-forward Neural Network (FFNN), and Adaptive Neuro-Fuzzy Inference System (ANFIS) for fruit classification tasks. The accuracy rate of the proposed technique surpasses that of SVM, FFNN, ANFIS, and CNN-LSTM. The results of the experiment showed that the developed model achieved an accuracy of 97.35%, F1-score of 94.91%, and Coefficient of Correlation (CoC) of 96.50 and RMSE of 11.50 respectively. Furthermore, the proposed approach contains relatively high computational momentum that will be further enhanced in the future.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.