Abstract: In order to meet the requirements of obstacle detection in the process of autonomous operation of agricultural machinery and solve the problem that visual detection is easily affected by operating environment, this paper proposes a field obstacle detection scheme that integrates millimeter-wave radar and camera information. Firstly, according to the radar cross section（RCS） value of the radar and the transformation matrix, the mapping of the radar point to the image pixel coordinate system is completed. Afterwards, the improved dark channel prior dehazing algorithm and the saliency-based enhancement algorithm proposed in this paper are used to complete the local image enhancement. Finally, the yolov4-tiny network is used for detection, and the decision-level fusion strategy is used to complete the data association between the millimeter-wave radar and the camera detection results, and comprehensively output information of the obstacles. The test results show that compared with the detection results of the visual detection, after incorporating the millimeter-wave radar information, on different datasets, the detection performance has been improved. Specifically, on the sunny day with dust dataset, R rises 16.4%, mAP rises 7.95%. On the foggy data set, R rises 17.7%, and mAP rises 6.63%. At the same time, the processing time for a single image of the algorithm after incorporating millimeter-wave radar information is about 148ms, which can meet the real-time requirements of agricultural machinery in the process of autonomous operation.