Hypothesis 1
Visual Distinction Exists
Healthy leaves and diseased leaves can be separated through average-image comparisons and montage exploration.
Validated by visualizer assetsReasoning Layer
H
Project Hypothesis and Validation
This section explains the core assumptions behind the disease classifier and shows how each one maps to visual analysis and model behavior.
AssumptionsValidation
Hypothesis 2
Classifier Can Reach High Accuracy
A CNN can achieve strong multi-class prediction performance for Healthy, Powdery, and Rust labels.
Validated by test metrics near 95%Hypothesis 3
Background Shift Impacts Reliability
Different image backgrounds may reduce model confidence and classification stability.
Monitored during live detector usageHypothesis 4
RGB Input Improves Consistency
RGB images are preferred for inference; non-RGB samples should be converted automatically before prediction.
Enforced in backend preprocessingOutcome Summary
- Visualizer confirms meaningful visual class differences.
- Performance metrics support strong generalization on test data.
- Detector flow preserves practical recommendations for disease handling.