A Comprehensive Analysis of Classic Machine Learning and Deep Learning Modeling for Breed Recognition of Bananas
Md. Ataur Rahman, M. Raihan, Mohammad Shorif Uddin, Mahady Hasan, Md. Tarek Habib
Abstract
Bananas are among the world’s most widely farmed, traded, and consumed fruits. Bananas contain several vitamins and minerals, including vitamin B6, vitamin C, magnesium, and manganese. Bananas are a very good source of vitamins and minerals in a densely populated and lower-middle-income country like Bangladesh since they are widely cultivated and available in urban and rural markets all over Bangladesh. Deep learning algorithms have shown considerable promise in recent years for image-based breed recognition tasks, encouraging researchers to look into their potential for breed-specific banana recognition. This research thoroughly investigates the use of deep learning and classical machine learning for banana breed recognition. Seven separate deep learning models were used to recognize bananas: VGG16, ResNet50, Mobile-Net, Inception-v3, MobileNetV3, Hybrid CNN-ViT, and a custom convolutional neural network (CNN), and five classical machine learning classification models: <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</i>-NN, SVM, Random Forest, Decision Tree, and Ensemble methods (SVM + Random Forest). These models were trained using images of several banana breeds to differentiate between them based on visual characteristics such as size, shape, color, texture, and skin pattern. The performance of each deep learning and classical machine learning classification model was thoroughly evaluated and tested. The customized CNN model has the highest accuracy of all models. We rigorously tested our models by comparing accuracy along with other key metrics such as recall, precision, and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</i><sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub>-score. In the cases of these metrics, the customized CNN model outperforms all other deep learning and classical machine learning classification models with an accuracy of 99.37%.