Optimizing Parameter Combinations in Convolutional Neural Networks using Improved Grey Wolf Optimizer for Rice Nutrient Deficiency Classification

Precision agriculture faces challenges such as population growth, shrinking arable land, and climate change, necessitating efficient plant nutrient management to enhance productivity while ensuring sustainability. Convolutional Neural Networks (CNNs) are crucial in detecting plant nutrient deficiencies. Still, their success depends on optimizing parameters such as the number of epochs, dropout rate, hidden layer size, learning rate, activation function, and optimizer. This research introduces a novel Improved Grey Wolf Optimizer (GWO) method for determining the optimal CNN parameter values. Unlike previous works, the proposed Improved GWO incorporates a novel adaptive parameter adjustment mechanism to enhance the exploration-exploitation balance, reduce the risk of premature convergence, and improve computational efficiency. To our knowledge, this is the first implementation of Improved GWO specifically tailored to optimize CNN parameters to detect nutrient deficiencies in rice plants. The research uses the Nutrient Deficiency Symptoms in Rice dataset from Kaggle. It compares the Improved GWO with other optimization methods, including the original GWO, Particle Swarm Optimization (PSO), Whale Optimization Algorithm (WOA), hybrid GWO-WOA, and the FOX optimizer. The results demonstrate that Improved GWO achieves optimal parameters with 8 epochs, a dropout rate of 0.1, a hidden layer size of 128, a learning rate of 0.002, the ReLU activation function, and the Adam optimizer. This method achieved a training accuracy of 86.77%, a test accuracy of 82.43%, and the fastest training time of 109.50 seconds. Compared with other methods, the original GWO and WOA exhibited slower convergence and lower accuracy, whereas PSO incurred higher computational costs. At the same time, the FOX optimizer and GWO-WOA were less efficient and stable than the Improved GWO. Through this innovative approach, the research successfully determines the best CNN parameter values using an effective and efficient optimization method. The Improved GWO method advances optimization techniques and provides a practical solution to support precision agriculture, particularly for detecting nutrient deficiencies in rice plants, thereby contributing to sustainable crop management.