Deep Learning-Based Classification of Bangladeshi Vegetables

Abstract

Vegetables are an important aspect of daily living for humans and animals, with around ten thousand plant species classified as vegetables globally. Roughly fifty species are economically relevant. Proper identification and classification of vegetables into subclasses or groupings are vital for better understanding and management. This studies the application of deep learning techniques for the classification of Bangladeshi native vegetables. Multiple convolutional neural network (CNN) designs, including InceptionV3, VGG19, and DenseNet121, to discover the most effective model for vegetable recognition. High-resolution photos of 24 different vegetable classes were utilized to train and evaluate each model, utilizing transfer learning with pre-trained weights from the ImageNet dataset. The models were fine-tuned with extra layers designed for our unique classification task. Among the models examined, DenseNet121 emerged as the best-performing algorithm, attaining an astounding accuracy of 99.45%. This model displayed great precision and recall across most vegetable classes, suggesting its robustness in handling varied visual attributes. The findings reveal significant potential for employing DenseNet121 in real-world agricultural applications, facilitating automated crop monitoring and quality assessment. The application of such deep learning models can boost agricultural output, assist effective resource management, and encourage sustainable practices, contributing to food security and economic development in rural regions. It aids in resource management, enhances economic benefits, maintains food security, supports sustainable practices, and is accessible to farmers, making it an essential tool for contemporary and efficient agriculture