Vegetable Classification using Deep Learning: Insights from Transfer Learning Models

Deep learning models for vegetable classification address a key demand in current agricultural and food sector activity. The significance of precisely recognizing and categorizing vegetables is multifaceted, ranging from agricultural process optimization to food safety and quality management. Deep learning models for vegetable classification meet a major demand in the agriculture and food industries, providing considerable gains in efficiency, production, and innovation. The present research aims to employ deep learning techniques to accurately categorize vegetable images. In this regard, the focus of this research is to discover whether conventional CNN architectures—such as VGG16, ResNet50, MobileNetV2, DenseNet121, VGG19, Xception will perform best in terms of accuracy and efficacy when applied to newly gathered image datasets. Vegetable Image Dataset of 21,000 images split into 15 categories. Our work into vegetable classification demonstrated the capabilities of six different deep learning models. With accuracy scores ranging from 98.23% for the Xception to an astonishing 99.97% for DenseNet121, each model brought its own capabilities to the table. In addition, the accuracy of ResNet50 was as high as 99.73%, second only to MobileNetV2 and DenseNet121 which achieved the results at a rate of 97.88% respectively Although, the models which stood out from others were VGG16 and vgg19 with accuracy as 99.53% and 99.13%, followed xception based model giving an accuracy of 98.23%.