SVM Classifier of Stored-Grain Insects Based on Grid Search

• Published in: Advances in Computer Science and Education Applications pp. 436 - 441
• Main Authors: Hongtao, Zhang, Shuping, Yin, Yuxia, Hu
• Format: Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg
• Series: Communications in Computer and Information Science
• Subjects: feature extraction; grid search; image recognition; Stored-grain insects; support vector machine
• ISBN: 9783642224553; 3642224555
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-642-22456-0_62

The detection of the stored-grain insects based on image recognition technology is high accuracy, cost-effective, high efficiency, no pollution and less labor. The classification of the stored-grain insects was multi-feature and multi-compound degree of various insects. The classifier design was critical for the detection system of the stored-grain insects. The optimal parameters C and g should be identified while using RBF kernels in the SVM classifier. The goal was to get the best cross-validation training model and improve the classification accuracy of the classifier. The grid search consisting of rough selection and fine selection was proposed to optimize parameters C and g by the recognition ratio of the training model. The optimal parameters were 30574 and 0.5743 after training, respectively. The fifteen species of the stored-grain insects that spoiled seriously in granary were automatically recognized by SVM classifier, and the correct identification ratio was over 94.67%. The experiment showed that it was practical and feasible.