TriggerNet: A Novel Explainable AI Framework for Red Palm Mite Detection and Multi-Model Comparison and Heuristic-Guided Annotation

Explainability is essential for agricultural AI: TriggerNet combines multiple explanation techniques (Grad-CAM, RISE, FullGrad, and TCAV) to show why AI models make disease detection decisions, making the technology more trustworthy for farmers and agricultural experts who need to understand and verify the system's reasoning.

Multiple AI models should be compared, not just deployed: The study evaluated six different deep learning models (CNN, EfficientNet, MobileNet, ViT, ResNet50, InceptionV3) plus three machine learning classifiers, finding that model comparison is crucial for selecting the most reliable tool—a responsible practice that prevents over-reliance on a single "black box" system.

Smart labeling reduces human bias in training data: The use of Snorkel with heuristic rules to label disease classes (Healthy, Yellow Spots, Reddish Bronzing, Silk Webbing) "reduced manual annotation time and improved dataset" quality, demonstrating how responsible AI development can minimize human annotation errors that could otherwise bias the model.

Visual explanations enable human oversight: By generating visual explanations showing which plant features trigger disease detection, the framework allows agricultural experts to catch potential errors—such as the model focusing on irrelevant image artifacts rather than actual disease symptoms—before deployment in real-world farming decisions.

Diverse training data across multiple plant species improves generalization: Testing across 11 different plant species helps ensure the AI system works reliably across varied agricultural contexts rather than only in narrow, controlled conditions, making it more responsibly deployable across different farming regions.