Real time efficiency is one of the bottleneck problems in the field of image processing, especially in the natural scene of the agricultural robot vision system. Nowadays superpixel segmentation algorithm was proposed as the high robustness to deal with the random uncertainty in natural scene. Simple linear iterative clustering(SLIC) has drawn much attention due to its outstanding performance in terms of accuracy， speed, antishadow and antihighlight. In this paper, by applying the Cauchy-Schwarz inequality, we derived a condition to leave unnecessary operations from the cluster inspection procedure. In the proposed algorithm, we reduced the redundant computation by using a robust inequality condition based on weighted L2 norm of pixel and cluster center representation. Then we put up with an advanced algorithm: FSLIC algorithm. We built a database with 2000 apple images in almost all natural conditions. Several kinds of extreme situations were chosen: high intensity of illumination light condition, low intensity of illumination backlight condition, uneven illumination of cloudy condition, adjacency and severe adhesion condition. The error rate curves of the insufficient segmentation, the hit rate curves of the boundary and execution time were analyzed with the 500 apple images; the GCE, FNR and FPR were detected with the 30 images in extreme condition. In the experimental results, it was confirmed that the GCE in Graphbased and FSLIC algorithm was reduced by 13% than BP algorithm, WT algorithm and SVM algorithm, the GCE in FSLIC algorithm was reduced by 19% than the traditional algrithms. The hit rate of the boundary in FSLIC algorithm was increased by 21.7% and the speed was 1.83 times than Graphbased algorithm.