Hybrid Feature Selection with Parallel Multi-Class Support Vector Machine for Land Use Classification

Land use classification in remote sensing is required in various applications like natural resource management, urban mapping and agriculture etc. Existing methods in the Land use classification which has the limitation of overfitting problem due to the improper feature selection in the method. In this research, the hybrid feature selection methods with Parallel Multi-Class Support Vector Machine (MSVM) is proposed to improve the land use classification performance. The UC Merced and AID datasets were applied to validate the performance of the hybrid feature selection method with the parallel MSVM method. The input images were applied in Histogram Equalization to enhance the image quality which removes the artifacts in the preprocessing stage. The Speeded Up Robust Feature (SURF), Local Ternary Pattern (LTP), Discrete Wavelet Transform (DWT) were applied for feature extraction. The extracted features are applied to hybrid feature selection of Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) method to select the relevant features. The hybrid feature selection method has the advantages of good convergence with higher efficiency in search analysis. The PSO model provided good search exploration to find better solution and GWO method has good convergence of local and global solution. The hybrid method has effective exploration and exploitation for the feature selection. The proposed hybrid features with the MSVM method have 99.15 % accuracy and the existing SVM has 94 % accuracy in land use classification. © 2022,International Journal of Intelligent Engineering and Systems. All Rights Reserved.