The Classification of Multispectral and Statistical Texture Data Using Data Mining Techniques

The real-time information for land use/land cover (LU/LC) data is very important for resource management, future prediction, and crops growth assessment. Although conventionally LU/LC data is collected through field survey but remote sensing data collection has its own importance due to time, accuracy and transparency factors. During the last decade, advancement in spaceborne multispectral data has proven to be beneficial over airborne data for land monitoring due to their increased spectral resolution. The objective of this research is to compare and analyze the five types (Fertile, Green pasture, Desert-rangeland, Bare and Sutlej-river land) of LU/LC multispectral data (five bands) acquired by multispectral radiometer (MSR5) and digital photographic data acquired from high resolution 10.1 megapixel Nikon camera. All experimentation has been performed using MaZda software version 4.6 with WEKA data mining tool version 3.6.12 on Intel® Core i3 processor 2.4 gigahertz (GHz) with the the 64-bit operating system. This research is conducted at The Islamia University of Bahawalpur province Punjab (Pakistan), located at 29°23′44″N and 71°41′1″E. For photographic data, image pre-processing techniques are applied, i.e., grayscale conversion, enhanced the contrast and sharpening procedure. Extract the 229 statistical texture features of the LU/LC data of each 512×512 image size. Three feature selection techniques fisher (F), the probability of error plus average correlation coefficient (POE+ACC) and mutual information (MI) are combined together (F+PA+M) and extract thirty most discriminant features out of 229 features space of each photographic image. For feature reduction, non-linear discriminant analysis (NDA) for photographic data (texture data) and linear discriminant analysis (LDA) for remote sensing data (multispectral data) have shown better clustering as compared to principal component analysis (PCA) and raw data analysis (RDA). Finally, we have employed different data mining classifiers namely, Artificial Neural Network (ANN), Random Forest (RF), Naive Bayes (NB) and J48 for classification. It is observed that artificial neural network (ANN: n class) is applied for training and testing by cross-validation (80-20) on these texture and multispectral data. It showed comparative better 91.332% accuracy for texture dataset and 96.40% for multispectral (MSR5) dataset respectively among all the employed classifiers.