A multispectral image based object detection approach in natural scene

Abstract

In this paper, we investigate the feasibility of blueberry detection based on multispectral image, and build robust classification models, which tolerate outdoor illumination changes and complicated background information. In the fruit growth, there are three stages, mature, near mature, and young. Base on the multispectral image and the normalized vegetation index NDVI, we construct a classifier to detect the different growth stage of the fruits by analyzing the color component and using the C4.5 algorithm. The experimental results show that the correct recognition rate of the mature, approximate mature and immature fruit are 74.58%, 79.91% and 86.52%, respectively, which demonstrate that the proposed classifier has a better recognition effect for immature fruits.