Performance Enhancement of Mixed Pixel Classification by Hybridization of Particle Swarm Optimization and Biogeography Based Optimization

Abstract

The images obtained from satellite gives the necessary data about the geographical condition and variations of the earth. It helps in the reduction of time while working in the field because images obtain from the satellite gives the relevant data which is full of quality and quantity. As the main focus of our work is on the classification of images obtain by satellites, satellite image classification is a strong technique which is used for the extraction of informative data. The term classification means a process of grouping pixels into meaningful classes. In each remotely detected image, an extensive number of mixed pixels are available. A mixed pixel is a picture element on behalf of an area engaged by in excess of one ground cover write. Essentially, there are two circumstances in which mixed pixels happen. The first case distresses the pixels that are situated at the edges of huge articles like rural fields, for example. The second case emerges when substances are imaged that are generally little contrasted with the spatial resolution of the scanner. newlineIn this work, firstly unclassified images are given as input then the segmentation is done by using Simple Linear Iterative Clustering (SLIC) and Density-Based Spatial Clustering (DBSCAN) clustering algorithms and the merge the superpixels. Feature extraction processes applied to the superpixel image and train the neural network on these features. Features are classified by using Random forest and Bagging method. In this work, classification is done by the hybrid PFCM algorithm which is a combination of PSO (Particle swarm optimization) algorithm and Fuzzy C-Mean algorithm and provides the local and global feature classification. Biogeography-Based Optimization is used for the classification of super mixed pixels. The performance evaluation of the proposed work is improved in terms of parameters that are TP rate, FP rate, Precision, Recall, F_measure, and accuracy. The overall accuracy of this work is 96.1 and RMSE is 0.032607. newline