Research On Construction Machinery And Material Identification Technology Based On Multi-Sensor Fusion

The research of intelligent unmanned loader has a great promotion effect on China’s intelligent manufacturing industry.In order to improve its recognition accuracy in complex scenes and advance its technological innovation,this study focuses on multi-sensor fusion of construction machinery and material recognition technology to solve this challenge.Specifically,the research covers the following aspects.First,in order to enhance the robustness of computer vision models in complex environments,various strategies are used to construct diverse datasets.These include using different devices to capture images from various angles,distances,sharpness,weather,and time of day.In addition,we also employ style migration and sliding window techniques to enhance the images and supplement the number of samples in complex environments and severe weather to further improve the model’s adaptability to different environments.Secondly,this paper proposes an improved construction machinery and material target detection algorithm based on YOLOv4-Tiny and VGG19.Aiming at the problem that the YOLOv4-Tiny basic network framework only adopts the form of layer-by-layer connection,which has insufficient feature extraction ability,this paper connects the lowlevel information with the high-level information in depth with the help of the network idea of VGG19,and applies the residual network with multi-scale feature pyramid network structure to output two improved prediction networks of different scales.Meanwhile,this paper also adds a CAM attention mechanism to the Neck network of CNN for secondary screening of feature information of the region of interest at each scale after the backbone network to filter redundant feature information,thus further improving the detection accuracy.The improved algorithm mAP performs well,achieving 93.98% and 73.42%recognition accuracy at IOUs of 0.5 and 0.75,respectively,which are 4.45% and 5.65%higher than before the improvement.Meanwhile,the FPS of the algorithm reached 155.2,and the comprehensive performance was excellent among the six types of comparison algorithms.Finally,this paper improves the recognition rate of unmanned loaders in complex environments by studying the fusion of millimeter wave radar and vision camera information.The paper considers that millimeter-wave radar has strong penetration ability in rainy and foggy weather and dim environment,so it is fused with the camera for perception.Specifically,in this paper,we first establish the spatial fusion model of millimeter wave radar and camera to achieve the unification of the spatial coordinates of the two sensors.Then we project the target acquired by the millimeter wave radar onto the image and generate the region of interest of the camera using the detection frame intersection and comparison model.Finally,the region of interest is identified using the improved YOLOv4-Tiny algorithm.In the experiments under low contrast environment and rain and fog environment,we found that the recognition rate of the machine vision-based solution is only about 80%,while the recognition rate reaches more than 90% when combined with millimeter wave radar.Through the verification of the real measurement data,the algorithm of this paper effectively improves the recognition accuracy of construction machinery and materials in complex scenes.