一种用于旋转不变分类识别的线性判别函数

提出利用线性判别函数，ｄ（ｘ）＝Ｗ＇ｘ，对轰炸机和战斗机两类目标进行旋转不变识别。每一目标有７２个旋转模式，其中９个为训练样本，６３个为测试样本，在每个目标的训练样本中随机取一个相减作为线性判别函数的系数（Ｗｉｊ）的初始值，然后利用误判的训练样本优化判别函数的权值以提高函数的正确识别率。为便于光电实现，线性判别函数的权值系数只有２、１、０、－１、－２五个值。计算机模拟表明利用此判别函数进行分类的正

A Linear Discrimination Function Used to Classify Objects with Rotation Invariance

A linear discrimination function (d(w, x)=w′x) is presented to classify two aircrafts (the bomber and the fighter) with rotation invariance. There are 72 rotation images (patterns) for each aircraft, in which 9 rotation patterns are used as the training set and the other 63 patterns are used as the test set. The original weights {wij} are obtained by one of the training pattern of the bomber subtracting one of the training pattern of the fighter. To improve the correctly recognizing rate, a modification of the correctly weights is carried out by the wrong recognized training patterns. In order to opto-electronically implement, the weights {wij} only have five values (-2, -1, 0, 1, 2). The results of the computer simulation state that the correctly recognizing rate reach to 95.8%. A opto-electronic hybrid system is presented to implement this linear discrimination function for recognizing the objects with the rotation invariance.