基于混沌优化K均值算法的马铃薯芽眼的快速分割

为提高芽眼分割的准确性,该文实现了基于混沌优化K均值算法的马铃薯芽眼的快速分割。K均值算法具有有效性及易于实现的优点,但是容易陷入局部最优值的缺点造成了其聚类结果的不准确。混沌系统由于其遍历性和不重复性,能够以较快的速度执行全局搜索。该文提出的算法的主要思想就是将混沌变量映射到K均值算法的变量中,用混沌变量代替其寻找全局最优值。分割试验结果表明:该文提出的算法,不仅在分割准确性上优于当下流行的K均值算法和模糊C均值算法,而且在运行时间上也更胜一筹,K均值算法和模糊C均值算法分割一幅马铃薯芽眼的图像所需的平均时间分别为2.895 5 s和3.556 4 s,而该文提出的算法仅需1.109 s。当聚类数大于3时,该文提出的算法在运行时间上受聚类数目的影响非常小,这一特点可用于其他一些适合聚类数大于3的农作物上。试验结果还表明,对于该文中的样本,最佳聚类数不宜超过3。最后,精度试验验证了该文提出的算法能够对样本中的马铃薯芽眼实现完全,无遗漏的分割,总的分割精度为98.87%,其中,对正常的马铃薯芽眼分割精度可达100%,对特殊马铃薯的芽眼分割精度为91.67%,总体分割精度较好。因此,该文提出的算法能够为后续种薯的自动切块提供参考。

Fast segmentation on potato buds with chaos optimization-based K-means algorithm

Abstract: Accurate segmentation of the potato buds is the premise of seed potatoes automatic cutting. In order to improve the accuracy of buds segmentation, the fast segmentation on potato buds with chaos optimization-based K -means algorithm was realized in this paper. K-means algorithm has the merits of effectiveness and easy to implement, while the problem of easily trapping into the local optima hinders its accuracy of clustering. Chaotic systems can carry out overall searches by chaotic variables at high speed with ergodicity and non-repetition. The main idea of the proposed algorithm was searching with chaotic variables by mapping them into the range of the variables in K-means algorithm and eventually achieved global optimization. Experimental results demonstrated that the proposed algorithm outperformed the state-of-the-art K-means and FCM (fuzzy C-means) algorithm, whether in segmentation accuracy or running time. The average running time for K-means and FCM algorithm to segment the buds in one image was 2.895 5 and 3.556 4 s, respectively, however, it only took 1.109 s with the proposed algorithm. And running time was less effected by clustering number when it was more than 3, namely, the proposed algorithm could provide a less running time when applied to other crops, which were suitable to be segmented into 3 or more clusters. Moreover, results also verified that the better selection of the clustering number for the samples in this paper was no more than 3. Finally, for the potatoes in buds segmentation precision experiment, the proposed algorithm could achieve a total segmentation precision of 98.87% and without any buds omission. Thereinto, the segmentation precision for normal potatoes was 100%, and that for special ones was 91.67%. Consequently, the fast segmentation of potato buds with the proposed algorithm could lay a solid foundation for future automatic cutting of seed potatoes.