In order to improve the efficiency and safety of underwater operation of deep-sea exploration robot, a map construction method combining inverse density gradient clustering and bilinear interpolation is proposed. The image of seabed environment collected by the robot is pre processed by graying, segmentation and denoising; Clustering the image pixels of the obstacle area; According to the structure size of the detection robot, the binary images after clustering are locally expanded by improved bilinear interpolation to obtain the final environment map. The results of map construction in two different environments show that, compared with the traditional Meanshift algorithm and bilinear interpolation algorithm, the image processing combined with inverse density gradient clustering and bilinear interpolation realizes the determination of infeasible regions in the map, with the average detection rate increasing by 26.1% and the average missing detection rate decreasing by 31.4%, which verifies the effectiveness of this method.