Image analysis technology is an important approaches for investigating the micro mechanisms of bed load transport. Research on improving the accuracy and reliability of this technology is extremely potential in sediment field. Combining the image recognize theory and bed load transport property, this paper proposed a more reliable approach to detect the bed load transport rate based on the image recognition technology. The images of sediment transport progress are captured by non-contact manner. Then, by means of image recognition technology, the sediment motion property parameters are obtained without affecting the flow and sediment motion. This is extremely valuable to keep the accuracy of the experiment data. During the progress, methods like Median Filter, convolution processing and et al., are adopted to enhance the image quantity. The adaptive threshold technique is used to segment the image. With the help of this technology, new method for calculating the real time grading and transport rate of bed load are proposed in the paper. Based on the abundant flume experiment data, the result of the computer recognition based on image recognition technology are contrasted with the artificial recognition in detail. It shows that the detect technology in this paper has a considerable accuracy and reliability. Particularly, when the bed load transport at a low intensity, the detect accuracy is extremely high. The accuracy of the technology could be correct by correction coefficient α when they transport at a high intensity. Meanwhile, the area distribution density of bed load on the bed could also be obtained. The technology in this paper improves the accuracy of measurement based on image recognition technology and supply a new method for the real-time bed load measurement. This makes sense to the development of the intensive study on non-uniform bed load motion.