Uneven material surface, likeliness to deform and other features of the flexible hygiene products bring great problems to the positioning and segmentation of key areas of the image in machine vision, thus affecting the detection results. In order to solve the problem of detecting dirty spots on complex shading flexible hygiene products, the work aims to propose a new detection algorithm to realize the rapid and accurate detection of complex shading flexible hygiene products. Firstly, the proposed detection algorithm pre-processed the collected template images and sample images, created a proper feature template from template images, matched and located sample images according to feature templates. Then, the key areas of the accurately located sample images were segmented. Finally, the difference shadow method was used to find out the defects in the key region of the sample. The proposed algorithm was implemented by halcon operator of MVtec company in Germany. Experimental result showed that the proposed algorithm could identify dirty spots greater than 0.04 mm2 on the surface of the cotton core. The average time cost was 100 ms, and the detection accuracy was 100%. Compared with traditional methods, the proposed method is robust, fast and able to meet the needs of industrial high-speed production.