论文部分内容阅读
提出核空间距离测度这一可分性判据.在核空间中计算两类样本点之间的距离,并以距离的大小评价子集的分类性能.使用顺序前进法作为搜索算法,在人造和真实的数据集上进行测试,文中的核空间距离测度可分性判据明显优于传统非核的可分性判据,优于或接近于W ang提出的核散布矩阵测度,并在运行时间上快一个数量级.将文中方法应用于胰腺内镜超声图像分类,取得较好分类结果.
The separability criterion of kernel space distance measure is proposed, the distance between two types of sample points is calculated in kernel space, and the classification performance of subsets is evaluated by the distance.Using the sequential approach as the search algorithm, Real dataset. The separability criterion of kernel space distance measure in this paper is obviously better than that of traditional non-kernel separability criterion, which is better than or close to the nuclear scatter matrix measure proposed by W ang. And in run time An order of magnitude faster.The method used in the text classification of pancreatic endoscopic ultrasound images, to obtain better classification results.