A Deep Learning-Based Framework for Disease Detection and Yield Enhancement in Five Major Crops

Precision agriculture is using more and more powerful computer methods to improve crop monitoring and produce management. This study introduces a new deep learning-based framework that combines disease detection in several crops with yield prediction. Its goal is to help with real-time agronomic decision-making. Using the publicly available "Top Agriculture Crop Disease India" dataset, a Hybrid CNN + LSTM + Attention model was made to classify diseases in five main crops: maize, potatoes, rice, sugarcane, and wheat. The suggested Hybrid model had an overall average classification accuracy of 86.9%, which was a statistically significant improvement above baseline CNN designs like ResNet50, EfficientNetB0, and VGG16. We used a CNN + XGBoost pipeline to estimate yield. It used deep features, illness severity scores, and environmental data that CNN extracted. This method gave an average validation accuracy of 93.84% over all five crops. The examination of feature importance indicated that deep features and illness severity are the main factors that affect yield results. The integrated system can make decisions in real time, with a processing time of less than 100 ms per sample. It is set up through a user-friendly online interface. Farmers may use the system to get useful information about how to manage diseases and get the most out of their crops. The results show that the framework could improve precision agriculture methods and make it easier to use with edge computing and multimodal data sources in the future.