| With the rapid development of science and technology,the development of traditional industrial technology combined with the Internet of Things,big data and information technology has promoted the development of intelligent manufacturing technology industry.The improvement of the utilization rate of production resources has liberated a large number of human demand.In the context of intelligent and modern production environment,the requirements for safe production are also high.By artificially monitoring the safety requirements in the production environment,its efficiency is difficult to meet the needs in the real scene.Aiming at the intelligent management of the intelligent construction site monitoring system,the problem of uneven illumination of the site,different scales of objects with different shooting angles and distances,and the problem of object occlusion and blurring of images,the target detection system helps solve the problem of potential safety hazards on the construction site.With the help of today’s popular deep learning target recognition technology and based on YOLOv4 algorithm,the target detection intelligent monitoring system developed has improved the accuracy of target recognition in the application environment of smart construction sites.The experiment selects the monitoring picture of the building construction site,enhances the overall picture of the monitoring image,optimizes the algorithm processing of the picture before detection,intercepts the position that needs to be identified in the monitoring picture according to the movement of the target to be detected with the predetermined size,and enhances the image in the case of dark light and foreign objects blocking.Aiming at the problem of multi-scale objects in the real shooting scene,the improved Mosaic data enhancement processing and the optimization processing of MSRCR image enhancement algorithm are proposed.The program can monitor the problems such as the dress standard of the detection staff for objects within the detection range in the complex environment of the real construction scene.The local attention mechanism module is designed to improve the training effect of the model by focusing on the area of attention.The problem of excessive local features caused by the global attention mechanism is avoided by using local attention features and spatial information.In order to improve the problem that small target information is difficult to extract and detail features are lost in the target detection algorithm,a priori condition for size estimation is introduced.Aiming at the problem of detecting target size at different shooting distances,the feature pyramid network model is improved to balance the scale of semantic feature changes between different levels,alleviate the problem of weakening feature extraction information,enhance the feature description ability for small targets,and alleviate the problem of feature loss.The experimental test results show that under the improved target recognition environment of image enhancement,the system has significantly improved the recognition accuracy for small targets at test points.The detection speed of a single image remains at about 47 FPS,with an average accuracy of 85.51%.Under the same experimental environment and test data set,the system has increased 6.73 percentage points.For the application environment of outdoor construction site,the application software of target detection is developed for the problems of construction worker detection,early warning of safe construction area,mutual occlusion between targets,small target scale,etc.The actual test and inspection prove that the improved Mosaic data enhanced MSRCR image processing technology and the improved feature pyramid network model for the detailed description of small targets in this study can improve the detection effect of targets with large shooting range and small detection targets at the construction site,and the recognition accuracy of the system for target detection has been improved.After testing,the system has a good application effect in the construction site. |
