Classification Of Butterfly Species Based On Venasi Using Support Vector Machine

Abstract

Butterfly is one of the most frequent object in lab activities of taxonomic courses with venation as a key feature of classification. This research is conducted to see whether this key feature of insect classification can be utilized to classify the type of butterfly with image of venation computationally. Classification process begins with the preprocessing and features extraction, then proceed with data sharing as much as K. Finally the training and testing are conducted using Linear and Non Linear SVM models. Features utilized on this research is a vector with the standard deviation as element of vector as much as quadsplit cutting. 120 schemes were tested for each value of K where K = 2,3 and 5. The highest accuracy attained where K = 2 is 97.05% for Linear SVM and 94.41% for Non-Linear SVM, where K = 5 is 97.64% for Linear SVM and 96.76% for Non-Linear SVM. Lastly, when K = 10 is 97.94% for Linear SVM and 97.94% for Non-Linear SVM. We found that Linear SVM accuracy value remained stable at 1024 cutting image, and  accuracy value decreases on Non Linear SVM. Also, The high value of the dimensions of the features can eliminate the non linear nature when is mapped to the kernel.