Abstract:In order to improve the terrain detection ability of unmanned platform in a complex environment and solve the problem of terrain recognition difficulty in small sample data, visual and touch fusion method for complex terrain detection on the unmanned platform is proposed. According to original broad learning, a multi-modal cascade feature nodes broad learning framework is established. First, extracting visual and tactile initial features, and then the fusion features are extracted. Next, the fusion feature matrix is classified by the broad learning classifier to obtain the classification results. Finally, experimental verification is carried out on the self-built data set named visual-touch terrain (V-T2). The results show that compared with the traditional fusion algorithms, the proposed fusion algorithm has good accuracy and robustness, and it provides an effective strategy for robot terrain detection.