Change Detection in Mangrove Forest Area Using Local Mutual Information

• मुख्य लेखक: Jahari, Mahirah
• स्वरूप: थीसिस
• भाषा: अंग्रेज़ी; अंग्रेज़ी
• प्रकाशित: 2010
• विषय: Mangrove forest; Crops - Growth; Forest biodiversity - Climatic factors
• ऑनलाइन पहुंच: http://psasir.upm.edu.my/id/eprint/19711/1/ITMA_2010_5_F.pdf

This thesis unveils the potential and utilization of similarity measure for forest change detection. A new simple similarity approach based on local mutual information is used to detect any significant changes in the image of forest areas. Point similarity measure is defined as a measure which is used to calculate the similarity of individual pixels. The basic idea of the proposed method is that any change pixel will be maximally dissimilar, i.e. the value of similarity of these pixels will be low. The method has been tested to detect and identify changes caused by plant growth and plant loss in four sub-areas of Matang Mangrove Forest, Perak. Image of SPOT 5 satellite taken from band 1, band 2,band 3, and band 4 with the resolution of 10meter dated on 5 August 2005 and 13 June 2007 has been used to test the method. It is then compared with the results of Principal Component Analysis 1 (PCA 1). The plant loss areas has been successfully identified as any pixel with the value of local mutual information less than and equals to zero. The method has been refined to accurately detect changes caused by the growth areas by thresholding the histogram of the average percentage of difference between joint probability and marginal probability. Results from the experiment showed that a threshold value of zero is the best threshold value to identify between changed and unchanged areas in all cases of the images. In overall, band 3 gives the best results of forest change detection compared to the other bands in all cases. Compared to the image differencing and normalized differenced vegetation index (NDVI), the proposed method not only can solve the problem on selecting the threshold value but also provides the highest percentage of successful classification at the fourth, second and first study area with the value of 95.07%, 89.47% and 87.66% respectively. From the results, it has been concluded that local mutual information is not only can be effectively used for change detection technique but also can be used to classify the plant growth and plant loss areas.