Symptom Driven Plant Diseases Classification

Objective

Developed a multi-modal classification system that diagnoses plant diseases using both image-based analysis and textual symptom descriptions generated by a language model.

Technologies Used

• Convolutional Neural Networks (VGG16, DenseNet, InceptionV3, Developed a Custom CNN)
• Multi-Modal Fusion
• Image Augmentation
• Gemini-Vision-Pro for symptom prediction

Key Achievements

• Achieved 95.5% training accuracy and 85% validation accuracy with DenseNet model after 10 epochs.
• Integrated image and text features to enhance disease diagnosis, improving classification accuracy over traditional image-based methods.
• Designed a custom CNN architecture with six 2D-convolutional layers, max-pooling, and fully connected layers for effective feature extraction and disease classification.
• Applied image augmentation techniques (scaling, rotation, flipping) to enhance training data variety.
• Utilized cosine similarity to evaluate the alignment between generated symptoms and reference texts, achieving high similarity scores for various disease classes.

Have a Peak at the Workflow

Results

• DenseNet performed the best among all models, achieving superior results in disease classification and symptom prediction.