CROP YIELD PREDICTION USING DEEP LEARNING

Abstract

The identification of an appropriate crop and fertilizer for every soil type is one of the key problems of modern agriculture. In the past, such decisions were based on experience, but technological progress today enables data-driven decision-making. This research recommends adopting a deep learning-based approach with a Lightweight Gradient Convolutional Neural Network (LGCNN) to classify images of soil into Black, Cinder, Laterite, Peat, and Yellow Soil. Training data are accessed from open databases and pre-processed using noise removal, segmentation, and feature extraction. Post-classification, soil types are mapped to pre-defined crop and fertilizer recommendations using a pattern-matching algorithm. LGCNN is more efficient compared to traditional models like Random Forest and is more accurate (98.96%). The system enhances precision farming using optimized resource allocation, improved crop yield, and sustainable agriculture. By combining machine learning and soil science, deep learning has the capability to significantly enhance farm output while not jeopardizing agricultural sustainability.