Enhanced Canopy Stress Detection In Vertical Farming Using Explainable AI

Vertical farming is revolutionizing agriculture by enabling the cultivation of crops in controlled environments, optimizing space usage, and reducing the need for traditional farming resources. However, the artificial conditions in vertical farms can induce stress in plants, affecting their growth and productivity. This paper presents a novel approach to detecting canopy stress in vertical farming using explainable AI techniques. The proposed method leverages deep learning models, specifically convolutional neural networks (CNNs), to analyze images captured by cameras and sensors within the vertical farm. By employing Grad-CAM (Gradient-weighted Class Activation Mapping), the model provides visual explanations for its predictions, enhancing transparency and trust in the AI system. The study utilizes the Agricola Moderna Healthy-Stress Dataset (AGMHS), comprising high-resolution RGB images of crops, to train and validate the model. The methodology includes pre-processing steps such as image resizing, noise removal, and contrast enhancement, followed by feature extraction using ResNet50 and dimensionality reduction with PCA. The classification of stress conditions is performed using CNNs, and the results are interpreted using Grad-CAM to highlight the critical regions influencing the model’s decisions. The proposed approach demonstrates high accuracy in detecting stress conditions, outperforming existing methods. The integration of explainable AI techniques ensures that stakeholders, including farmers and researchers, can understand and trust the model’s predictions, ultimately contributing to more efficient and sustainable vertical farming practices.