OPTIMIZATION OF THE SVM REGULARIZATION PARAMETER C IN MATLAB FOR THE OBJECT-BASED CLASSIFICATION OF HIGH VALUE CROPS USING LIDAR DATA AND ORTHOPHOTO IN BUTUAN CITY, PHILIPPINES

摘要

This paper describes the processing methods used for the detailed resource mapping of different high value crops in Butuan City, Philippines. The proposed methodology utilizes object-based image analysis and the use of optimal features from LiDAR data and Orthophoto. Classification of the image-objects was done by developing rule sets in eCognition. LiDAR data was used to create a Normalized Digital Surface Model (nDSM) and a LiDAR intensity layer. The nDSM and LiDAR intensity layers were then paired with Orthophotos and were segmented using eCognition for feature extraction. Several features from the LiDAR data and Orthophotos were used in the development of rulesets for classification. Generally, classes of objects can't be separated by simple thresholds from different features making it difficult to develop a rule set. To address this problem, the image-objects were subjected to a supervised learning algorithm. Among the machine learning algorithms. Support Vector Machine learning has recently received a lot of attention and the number of works utilizing this technique continues to increase. SVMs have gained popularity because of their ability to generalize well given a limited number of training samples. However, SVMs also suffer from parameter assignment issues that can significantly affect the classification results. More specifically, the regularization parameter C in linear SVM has to be optimized through cross validation to increase the overall accuracy. After performing the segmentation in eCognition, the optimization procedure as well as the extraction of the equations of the hyper-planes was done in Matlab. The optimization process can be time-consuming. To resolve this, parallel computing is employed for the cross validation process which significantly speeds up the process. The learned hyper-planes separating one class from another in the multidimensional feature space can be thought of as super-features which were then used in developing the classifier rule set in eCognition. In this study, we report an overall classification accuracy of around 95%. Seven features from the segmented LiDAR nDSM and intensity layers were used; area, roundness, compactness, height, height standard deviation, asymmetry and intensity. Eight features from the segmented Orthophotos were used; two features ( a* and b*) from C1ELAB color space, three features (x, y and z) from CIEXYZ color space, two features (first, and second coordinate) from one-dimensional scalar constancy, and one feature called RGB Intensity. We also show the different feature-space plots that have driven the proponents to use the aforementioned features for all the different classes.