Kemi-creates/CassavaScan_AI

Cassava disease detector AI model for African farmers. # CassavaScan_AI Cassava disease detector AI model for African farmers. ## The Problem Cassava is one of the most cultivated crops across Africa. Over 800 million people depend on it, but diseases like Cassava Mosaic Disease and Brown Streak Disease can silently destroy entire harvests before a farmer notices anything is wrong. The average farmer in Uganda, Nigeria, or Tanzania cannot afford an agricultural expert. There are no labs nearby. By the time a diagnosis is made locally, these diseases would have already caused severe crop yield loss. That loss isn't just income, it's also loss of food. ## The Solution CassavaScan AI is a web app that allows a farmer or agricultural extension worker to snap a cassava leaf and receive an **instant AI-powered diagnosis** that identifies which of 4 major diseases are present, or confirming the plant is healthy. This would help with early detection of these diseases, thereby, reducing crop yield loss. ## Live Demo > **Try CassavaScan AI on Hugging Face Spaces** > **Short Demo Video** ## Model Performance | Metric | Value | |---|---| | Architecture | MobileNetV2 (Transfer Learning) | | Training Images | 5,656 | | Validation Accuracy | 75.3% | | Test Classes | 5 | | Training Platform | Google Colab T4 GPU | ### Confusion Matrix ### Per-Class Performance | Disease | Detection Accuracy | |---|---| | Cassava Bacteria Blight (CBB) | 54% | | Cassava Brown Streak Disease (CBSD) | 72% | | Cassava Green Mottle (CGM) | 60% | | Cassava Mosaic Disease (CMD) | 86% | | Healthy | 79% | The model performs strongest on CMD and Healthy, the two most common real-world cases. Minority class detection is an area that is acknowledged to need more improvements, requiring more labeled training data. ## Dataset * **Source:** TensorFlow Datasets (`cassava`) * **Origin:** Makerere AI Lab, Makerere University, Uganda * **Images:** 9,430 real field photos crowdsourced from Ugandan farmers * **Why this dataset:** Captured under real Afr …