Crop Disease Detection and Treatment Recommendation (DL + ImageDectector)
Our project aims to revolutionize agriculture by providing accessible support to farmers, particularly those in remote areas, contributing to increased crop yields and reduced losses. In our exploration of Machine Learning(ML) models for crop disease detection, we rigorously tested several deep CNN architectures, mainly Generic CNN, ResNet50, EfficientNet V2, VGG16, DenseNet121, and our Hybrid Model (VGG16 + DenseNet121). The Generic CNN served as our baseline, achieving a commendable validation accuracy of 62%, providing insights into the initial predictive capacity of our system. ResNet50, with skip connections to address vanishing gradient issues, achieved an impressive validation accuracy of 74.57%. EfficientNetV2, incorporating external images, demonstrated efficiency with a validation accuracy of 73.07% .VGG16 and DenseNet121, individually assessed, showcased robust capabilities with accuracies of 76% and 79.82%, respectively. The subsequent Hybrid Model (VGG16 + DenseNet121) achieved a remarkable validation accuracy of 86.53%, outperforming individual models on the accuracy metric. Early stopping tactics optimized efficiency by halting training when no increase in accuracy was discerned. Our project contributes to precision agriculture by providing a reliable tool for early disease detection and treatment recommendations. The hybrid model’s success emphasizes the efficacy of leveraging complementary strengths in a unified architecture for crop disease detection.
This project was created with a timespan of 1.5 years, this includes everything from initial phases to final completion of the project.
It includes:
- High Level Design Document
- Low level Design Document
- Project Requirement Specification
- Project Final Report (Phase I + II)
- Monthly progression reports/presentations
- Design Diagrams (Architecture, Class, Deployment, Interface) with RAW files
- Literature Survey + Research Papers Applied
- Model Explanation
- Implementation Document
- Comparison Table
- Final working code
- Code Documentation
- DEMO Video + Results Graphs
- Crop Disease Dataset (17+10 classes with over 200 images)
This project has been heavily documented all the way from design phase to the implementation and testing phase. The project was divided into 2 phases:
- Phase 1: Design Phase
- Phase 2: Implementation Phase
450+ Pages of documentation