This paper proposes a Lepidoptera insect image recognition method based on extracting image features using superpixels segmentation, encoding the features with Locality-constrained Linear Coding (LLC), aggregating codes with max pooling, and then classifying them with classification and regression tree (CART). This method used the natural scale color patterns on the insect wings as the basis for recognition, which can avoid the complicated chemical processing needed for venation based recognition. The method is tested in a dataset including 579 image samples from ten species of Lepidoptera species, and the recognition error rate is below 5%. The method also exhibits good performance with respect to time cost. The experimental results suggest that on the task of recognizing Lepidoptera species, the proposed method has state-of-the-art performance with high efficiency.
How to translate text using browser tools
1 July 2013
Using CART and LLC for image recognition of Lepidoptera
Le-qing Zhu,
Zhen Zhang
ACCESS THE FULL ARTICLE
The Pan-Pacific Entomologist
Vol. 89 • No. 3
July 2013
Vol. 89 • No. 3
July 2013
coding
Decision tree
image processing
max pooling
scale pattern recognition
super pixels