TOWARDS ECOLOGICAL SUSTAINABILITY: HARVEST PREDICTION IN AGRIVOLTAIC CHILI FARMING WITH CNN TRANSFER LEARNING

Agrivoltaic systems, which integrate agricultural production with solar energy generation, present a promising approach to ecological sustainability. This study focuses on predicting chili harvests within an agrivoltaic setup using Convolutional Neural Networks (CNN) with transfer learning. Accurate yield prediction is vital for optimizing both agricultural output and energy generation. The study evaluates three pre-trained CNN models—EfficientNetV2L, EfficientNetV2M, and ResNet 50—fine-tuned with specific agrivoltaic data. The experimental setup includes a solar-powered greenhouse with IoT-controlled micro-climate management to ensure optimal growing conditions. The models were selected based on their high accuracy in Keras applications, with EfficientNetV2L and ResNet 50 achieving 100% accuracy, and EfficientNetV2M reaching 96% in chili crop counting. The results show quick convergence during training and validation, indicating effective model learning. The study also includes a life cycle analysis (LCA), confirming that using photovoltaic systems as a substitute for conventional energy sources is environmentally sustainable. Overall, this research demonstrates that CNN transfer learning is highly effective for crop counting and resource management, contributing to sustainable agrivoltaic farming and highlighting the potential of advanced AI techniques in agriculture.