IMPROVED CAT SWARM OPTIMIZATION WITH ENSEMBLE LEARNING FOR EFFICIENT CROP AND WEED CLASSIFICATION

Abstract

In recent days, Image processing has been extensively utilized in several fields; its applications appear in agriculture and clinical fields. To classify the weed still physical power is employed in various parts of the world. Physical power has been applied for weed identification in several parts of the world. After that, several approaches for detecting weeds without human intervention appeared and they could not reach the public because of shortage in accuracy. The large space requirements and higher computation time are considered issues in the current system. It also has issues in the feature selection mechanism for attaining superior features to progress the performance of learning. To deal with the issues of the current system mentioned above in the proposed system, the Improved Cat Swarm Optimization (ICSO) with EL-MCNN, Weighted Support Vector Machine (WSVM), and Adaptive Neuro-Fuzzy Inference System (ANFIS) algorithms are introduced. The proposed system includes main phases like pre-processing, image sharpening, feature extraction, feature selection, and classification. In the pre-processing step, the dynamically weighted median filtering algorithm is introduced to take off the noise produced in images efficiently. Image sharpening is done by using a piecewise regression model which has been utilized to increase the quality of images. Feature extraction is performed using quad histogram and Gray Level Co-occurrence Matrix (GLCM) for the utilization of extracting the Edge Orientation Histogram (EOH) feature and shape features effectively. The feature selection is completed by using the ICSO algorithm through the local search optimum and average inertia weight values. Then the EL-MCNN, WSVM, and ANFIS algorithms are proposed to classify the samples into crop, weed, and background accurately. The performance metrics are considered in terms of precision, recall, specificity and, f-measure which are evaluated using existing ensemble MCNN and proposed EL- MCNN, WSVM and, ANFIS algorithms.