In this paper, an improved canny edge detection algorithm was represented to solve traditional edge detection algorithm in rice edge detection which had low precision and noise sensitive. Firstly, nonlinear diffusion filter was used to wipe of noise efficiently and kept the edge information of the image. Secondly, gradient calculation of pixel diagonal direction was considered in the calculation of neighborhood gradient amplitude which further repressed the impact of noise. Thirdly, using average interclass variance could self-adaptively calculate the double thresholds for different images. The results of the experiment indicate that the improved algorithm not only can be applied to rice image edge detection but also has a better accuracy and precision in the edge detection.