PSO and K-means-based semantic segmentation toward agricultural products

Abstract Accurately segment agriculture products semantically is an indispensable technique in computer vision. YCbcr space separates brightness from color difference and has a linear conversion relationship with RGB space. In this work, particle swarm and K-means hybrid clustering are leveraged to segment agriculture product images in YCbcr color space, which can effectively alleviate the influence of low illumination and shadow, and achieve a balance between global solution search ability and convergence speed. The segmentation stability and accuracy of the proposed method are better than the traditional PSO clustering segmentation method. It can accurately and effectively segment the agriculture product images in various complex environments, which can facilitate the automatic agriculture product picking robot. In order to solve for K-means and the problem of poor segmentation effect of super green RGB image commonly used in agricultural images, an image segmentation algorithm based on particle swarm and K-means algorithm is proposed. K-means algorithm can dynamically cluster, but the results of this algorithm are easily affected by the number of clustering centers K and the initial clustering center, and are not stable. Therefore, this algorithm is not suitable for super green RGB agricultural images with less green information, and its segmentation effect is poor. The improved PSO – K algorithm solves the shortcomings of PSO’s slow convergence speed and K- means algorithm’s correlation with the initial clustering center. Besides, the improved PSO-K algorithm is an effective agricultural image segmentation method, which can accurately segment the target from different types of agricultural images after gray processing with super green feature RGB.