Multi-Feature Patch-Based Segmentation Technique in the Gray-Centered RGB Color Space for Improved Apple Target Recognition

In the vision system of apple-picking robots, main challenge is to rapidly and accurately identify apple targets with varying halation shadows on their surfaces. To solve this problem, study proposes a novel, multi-feature, patch-based image segmentation technique using gray-centered red-green-blue (RGB) color space. The developed method presents multi-feature selection process, which eliminates effect in images. By exploring all features image, including shadows, RGB space, proposed algorithm, generalization K-means clustering provides an efficient target result. tested 240 It offered average accuracy rate 98.79%, recall 99.91%, F1 measure 99.35%, false positive 0.04%, negative 1.18%. Compared classical methods conventional algorithms, as well popular deep-learning can perform high efficiency guide robotic harvesting.