In order to realize the unsupervised segmentation of small magnetic ring surface micro-defect image and improve the segmentation accuracy and computational efficiency,an adaptive threshold segmentation method based on an improved 2D Gabor filter bank is proposed.Firstly,a multi-scale and multi-direction Gabor filter is used to filter and de-noising the defect image,so as to suppress noise pollution within the target region and background region and enhance the difference between areas.Then,through the analysis of the statistical characteristics of the processed the gray of the image,based on the gray mean value and variance of the defect image,the gray threshold calculation formula is constructed,and an adaptive segmentation of the small magnetic ring surface fine defects images was realized.The experimental results show that the proposed algorithm segments the defects and reduces the noise accurately.It outperforms the traditional methods in terms of segmentation accuracy and computational efficiency,such as selective iterative,OTSU,maximum entropy and other methods,and it can be implemented in real time on the advanced SEED-DVS8168 platform,verifying the feasibility and real-time performance of this algorithm.