The invention relates to a dehaired cashmere length detection method based on a maximum interclass variance. The method comprises the following steps that: collecting the RGB (Red, Green and Blue) image of manual arrangement dehaired cashmere; preprocessing the RGB image; determining an image binarization threshold value; according to the image binarization threshold value, converting the preprocessed image into a binary image; carrying out edge curve extraction on the binary image to obtain an edge curve h(x); carrying out curve fitting on the edge curve h(x) to obtain a fit curve z(x); and according to the fit curve z(x), calculating the average length of the dehaired cashmere. By use of the method, the problem in an existing dehaired cashmere length manual arrangement measurement methodthat manual point tracing and plotting has manual measurement errors can be solved, in addition, the method can determine the image binarization threshold value according to the difference of image collection equipment and environment so as to obtain effective dehaired cashmere length, the constraint condition of the prior art for the image collection equipment and environment is avoided, and therefore, the detection method is more flexible in use.