Diverse Pumpkin Varieties in India: A Four-Class Classification Study with CNN and Deep Belief Networks Featuring Arka Suryamukhi, Pusa Vishwas, Pusa Akshay, and Pusa Vikas

This study delves into the classification of pumpkin varieties prevalent in India, focusing on four distinct classes: Pusa Vishwas, Pusa Akshay, and Pusa Vikas are four brave men. Using Convolutional Neural Networks (CNNs) and Deep Belief Networks (DBNs), a carefully made set of 11,200 high-quality pictures for thorough study. The data set went through tough cleaning, which included making everything equal, adding more pictures, and fixing the colors. This was done to make it strong and have many types of info in it. After dividing the training data into 80% for learning, 10% to check accuracy, and another 10% for testing. The models were trained using a method called backpropagation with categorical cross-entropy loss optimization while also adding dropout as regularization so it doesn't over fit too much. The models got very good results, being 96.7^% right overall. They did a great job with precision, recall, and F1-scores checking them up too. Looking at how it works compared to other ways, it found that the new method using CNN and DBN is better at correctly sorting different types of pumpkins. The study found that the model can be used in many ways, is scalable, and could help with picking food types on farms. This automated way helps make better decisions to manage crops well and leads to improvements in farm tech.