Novel Swarm Intelligence Optimized Quantum CNN Framework for Accurate Multi Class Soil Image Classification

Soil image classification aids precision agriculture, but current methods often fail in multi-class identification due to poor features and parameter tuning. This research presents a framework integrating Deep Learning (DL) and swarm intelligence optimization to enhance classification performance. A dataset of 1,378 distinct soil images from Kaggle was evaluated, featuring various soil categories including alluvial, black, cinder, clay, laterite, peat, red, and yellow soils. The proposed Local Gabor Rank Pattern with Quantum Convolutional Neural Network (LGRP + Q-CNN) extracts micro-texture features via LGRP and global semantic patterns via Q-CNN, while quantum computing improves feature encoding, accelerates optimization, and enhances classification robustness. In preprocessing, images were resized, noise removed with a Gaussian filter, and contrast enhanced. The Ring Toss Game-Based Algorithm (RTBA) optimized hyperparameters including learning rate, batch size, number of filters, and kernel size, aiming to maximize validation accuracy as the fitness function. The optimized LGRP + Q-CNN model, trained in Python 3.10, achieves superior performance over traditional DL and feature-based models with 98.5% accuracy, 96.8% precision, 96.65% recall, and 96.72% F1-score, supported by statistical analysis using Friedman and paired t-tests.The findings demonstrate that combining DL with swarm optimization offers a robust and accurate method for multi-class soil image classification.