Abstract: With the background of the rapid development of 3G communication technology, and the high popularizing rate of rural cell phones, this paper analyses the features of agricultural knowledge videos, which are currently in huge demand, but too big and too rough, and proposes the Method of Video Segmentation for vegetable disease based on self-adaptive dual thresholds. First, transform and quantify RGB video images based on HSV color space, and then divide the video with dual thresholds, which employs the self-adaptive threshold value to gain the best video segmentation threshold, and determine the abrupt and gradual transition by comparing the value of difference between frames and the threshold. Second, it's worth mentioning that in the measurement of frames, this method takes the segmentation strategy of different weighting factors, which excludes the influence of non-related content and makes the segmentation more precise. Finally, for the false-detected frames, because of the errors of threshold, an image similarity check method is suggested to revise the videos after segmentation, and combine the over-segmented ones, and finally output videos with continuous contents. This paper conducted an experiment on three videos about vegetable diseases, and compared the results in both histogram and dual thresholds, and reflected the effects with the rates of recall and precision. The experiment showed that the overall design of method of video segmentation for vegetable disease based on self-adaptive dual thresholds is able to segment videos quickly with a recall rate larger than 95% and a precision rate of 100%. Especially after the recheck in a similarity method, false-detection is almost distinguished. Compared to two other methods, it enhances the precision with similar segmentation time. So, this segmentation method ensures the precision rate without compromising the segmentation rate, and satisfies the information demands of specialized and individualized knowledge, and solves the conflicts of image quality and time, quantity and expense of net flow for farmers when they watch agricultural knowledge videos through their 3G mobile phones. Also, the method can provide a reference for other video segmentation of agricultural knowledge.