The invention discloses an improved image edge detection method. The method comprises the following steps of S1, using adaptive smooth filtering and smoothly inputting an image; S2, using the first order gradient template of a Sobel operator for reference and extending to the first order gradient templates in four directions of a horizontal direction, a vertical direction, a 45 degree and a 135 degree, carrying out convolution on the filtered image and acquiring first order gradient components at the four directions, and acquiring gradient amplitudes and gradient angles; and S3, using an Otsualgorithm to calculate a threshold, and combining a connection analysis method to determine the final edge of the image. In the invention, a traditional Canny edge detection algorithm is improved; firstly, the adaptive smooth filtering is used to replace Gaussian filtering in an original algorithm; then, the templates of the four directions of the horizontal direction, the vertical direction, the45 degree and the 135 degree are used to calculate a image gradient, noise sensitiveness is reduced and edge positioning precision is increased; after a non-maximal suppression step, the Otsu algorithm is introduced, and according to image gray level information, high and low thresholds are adaptively generated.