Compartive Analysis On Multispectral Images Based On Fuzzy Approach

• Main Author: Taufik, Afirah
• Format: Thesis
• Language: English; English
• Published: 2020
• Subjects: G Geography (General)
• Online Access: http://eprints.utem.edu.my/id/eprint/25439/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=119829

In particular, the vital information for classification can get lost due to a vast amount of data flow in the classification of remotely-sensed images. Nonetheless, existing techniques used for classifying mixed pixels in remotely-sensed imagery are not too efficient due to the homogenous category. In this study, the information of multispectral data of Landsat 8 is extracted to the three indices are used in this study are to represent of three categories; vegetation, non-vegetation and water body, are normalized difference vegetation indices (NDVI), normalized difference built-up indices (NDBI), and normalized difference water indices (NDWI). The indices are described as the input data for the methods of classification. In the present study, the fuzzy approach methods are developed and tested for a classification land cover mapping. An investigation is conducted based on a comparative study between fuzzy c-means, fuzzy supervised (adaptive neuro-fuzzy inference system) and other unsupervised methods, such as k-means. The evaluation of classification approaches on the ability to classify land cover classes with per-pixel digital image classification techniques is based on the user, producer and overall accuracy and kappa coefficient. For imbalance image datasets, the Klang and Krai image are compared to observe the distribution of data affect into user's accuracy (UA) and producer's accuracy (PA). For Klang data, the results show that the method of FCM performs better for UA in the non-vegetation class and PA in the vegetation class, with a percentage of 95.2% and 98.7% respectively. For Krai data, the method of FCM performs better for UA in the vegetation class and PA in the water class with a percentage of 98% and 99% respectively. In future work, more indices and category of classes can be considered to deal with the multispectral data for Landsat 8.