为在生产现场及时检测出发动机缸盖气门座圈未被压装到位的情况，采用蚁群聚类算法对状态各异的缸盖气门座圈缝隙图像进行合理归类，并在聚类基础上推演相关支持向量机(SVM)模型实现检测的早期分流，从而提高现场运行速度与检测判别性能。首先，分析缸盖座圈缝隙图像差异，确定体现检测图像宏观特征的观测点，采用聚类数目未知时的蚁群聚类算法确定基本分类估计。然后，对不同来源的缸盖所表现出的差别变异，通过聚类数目已知时的蚁群聚类算法推演例外的新类，借此自动覆盖所有可能出现的多种情况，以便在测试快速判别检测样本的归属。另外，借助聚类结果提示选取区分度高的样本，确定相关的SVM模型，将明确属于合格的多数缸盖从待检测队列中分流出去，从而显著提高整体检测的速度。实际测试表明，采用本方法所构成的检测装置错判率误差小于0.5%，不可容忍错判率为零，可满足工业生产现场检测的需要。该方法对状态各异以及多变图像的处理具有参考价值。

To detect if the alloy steel socket sleeve is built in a correct location, an incorporated method for image clustering via the ant colony optimization algorithm and the character differentiation model based on the Support Vector Machine(SVM) was proposed to improve the detection speed and estimation reliability . Firstly,the observation stamps about all of the images were determined according to the differential analysis between each type of the gap images. Then, the first turn of ant colony algorithm for the clustering number unknown was conducted to ascertain the number of basic sorts, and the second turn of ant colony algorithm for the clustering number known was performed to conclude the exceptional sort for the gap images from the another source. Therefore, the every possible states could be covered with the developed differential program. Furthermore, an advanced distributary method based on SVM model was adopted to increase the estimation speed etc. The experiment results in production line indicate that the mis-estimated error ratio is less than 0.5% and the irretrievable error is near zero. The instrumentation equipped with above technique can meet the need of industry application well.