COMPARATIVE ANALYSIS OF STANDARD AND ENHANCED SWARM INTELLIGENT DEEP BELIEF NETWORK ON MAIZE DISEASE DETECTION

Abstract

Deep Belief Network (DBN) and some other Artificial Neural Network (ANN) have been used for the detection and classification of diseases in plant but have been known for over-fitting problem. This has often times affected the accuracy of the system with high false positive rate. Particle Swarm Optimization (PSO) technique, has been used to enhance some deep learning techniques but still suffers stagnation in local optima and high computational cost mainly due to the large search space. Hence, there is need for improvement to develop a more efficient model for plant diseases detection and classification using Deep Belief Network with Improved Particle Swarm Optimization (DBN-IPSO). Images of healthy and unhealthy 3852 maize plant samples were acquired from https:\\www.kaggle.com. The acquired data were pre-processed, leaf colour images were converted to grayscale, cropped and contrast-enhanced with local histogram equalization. IPSO was formulated by incorporating modified acceleration coefficient and velocity component into the standard PSO to avoid premature convergence and imbalance between exploration and exploitation stages. The PSO was applied to DBN to select its optimal weight parameters. The system was implemented using a language app designer, trained and tested using k-fold cross-validation method, where k is 10-fold. The performance of the developed DBN-IPSO technique was compared with existing DBN-PSO and DBN using False Positive Rate (FPR), sensitivity, specificity, overall accuracy and computation time. The result shows that the developed technique improves the accuracy and efficiency of the system, making it a promising approach for automated plant disease diagnosis and surveillance in agriculture.

Keywords

• modified acceleration coefficient
• deep belief network
• particle swarm optimization
• premature convergence
• local optima stagnation