A pavement disease detection system with improved SSD (Single Shot MultiBox Detector) model is proposed to solve the problem of high dependence on high-performance servers. The detection algorithm adopts the method of first identifying and then classifying, detecting the various types of diseases on the road surface through the SSD model, performing the image gradient operator algorithm on the existing disease types, and then classifying the detected diseases by the VGG model. The hardware system adopts the Jetson-nano board as the main control board of the disease detection system, acquires the video data stream, loads the improved model, and loads the model by using the go language to load the tensorflow framework to improve the system running speed. The practical application results show that the improved road surface disease detection algorithm of SSD model realizes various types of high-precision disease detection; the accuracy of road surface disease detection reaches 88.26%; the average frame rate of detection system processing video stream is 15 fps, and the road surface disease detection system is realized. The practical application effectively reduces the cost of the detection system application and has a wide range of practical application values.

CICTP