| In the context of carbon neutrality and carbon peak,strengthening clean and efficient use of coal is not only a national strategy,but also the only way to high-quality coal development.The separation of coal gangue is an important way to improve clean and efficient utilization of coal.The development of gangue separation robot can realize automatic identification and sorting of coal and gangue,realize efficient separation of coal gangue,promote the development of intelligent coal mine and strengthen clean and efficient utilization of coal.At present,the accuracy and efficiency of coal waste identification algorithm need to be improved.Due to the absence of multi-scene coal waste data set,the detection process is susceptible to noise and motion ambiguity,and the automatic sorting method needs further research.Therefore,in combination with the theoretical knowledge of mining machinery,robotics,mineral processing,deep learning,image recognition,computer science,optics and advanced mathematics,this dissertation fully considers the real working scene of coal waste separation and deep learning algorithm,proposes a fast detection algorithm suitable for coal waste identification and a sorting method to block the target coal waste,and develops a robot experimental platform for waste separation for experimental verification.The main research contents and conclusions are as follows:(1)Construct multi-scene coal gangue data set.Aiming at the problem of coal gangue data set not yet disclosed,this paper collects coal gangue image samples in various scenarios,including images of different sizes,different illumination,mutual occlusion and large quantity of coal gangue,laboratory simulation environment and coal gangue images on site of coal preparation plant,etc.This dissertation studies an automatic image labeling method of coal gangue.A significant data enhancement method is proposed to realize the effectiveness expansion of coal gangue data set,and finally the efficient construction of coal gangue data set is realized.(2)Research on deblurring algorithm and noise reduction algorithm of coal gangue image.The real-time image detection of coal gangue is affected by noise,motion blur and other factors,which reduces the image quality and affects the accuracy of coal gangue detection.Therefore,image preprocessing is needed to improve the image quality.This dissertation proposes a fuzzy image de-blurring algorithm based on generation adduction network,a wavelet threshold and a bilateral filtering fusion image de-noising algorithm.Experiments show that the proposed algorithms can achieve rapid de-blurring and denoising of coal gangue image.Compared with other algorithms,the proposed algorithms have the largest evaluation index PSNR and SSIM values and have the best effect,improving the quality of coal gangue image.(3)Research on multi-scale fusion algorithm for light briquette gangue recognition.Aiming at the shortcomings of the current coal gangue recognition algorithm,such as low precision,slow speed,large memory consumption and poor detection effect of small size,a multi-scale fusion lightweight coal gangue recognition algorithm model is established.The model adopts Mobile Netv3 lightweight feature extraction network and replaces all SENet modules in the feature extraction network with SKNet.By adding shallow detection scales to the detection layer,a four-scale detection structure is formed.By adding space pyramid pool(SPP)module,a lightweight coal gangue detection algorithm is obtained,which can effectively and quickly detect coal gangue in small targets and overlapping targets.The m AP reaches 98.97%,the detection speed reaches 92 fps,and the memory consumption is only 11.7M.Compared with the original YOLOv3 algorithm,m AP is increased by 0.37%,fps is increased by 119.04%,and memory is only 1/21 of the original,realizing the real-time recognition of coal and gangue.(4)Research on sorting methods of coal gangue.The forward and inverse kinematics of the robot are analyzed by using D-H method for modeling,and the analytical equations of the forward and inverse kinematics are verified by using MATLAB simulation.For the trajectory planning of the robot,the trajectory planning method of 3rd plus 5th polynomial3-5 is proposed when gangue is not blocked,and the 3-5-3 trajectory planning method of3 rd plus 5th and 3rd plus 3rd is proposed when gangue is blocked,which can ensure the stability of the robot griper when grasping gangue,and the acceleration in the grasping process is not zero.And put forward the sorting target selection method,comprehensive consideration of gangue to the distance,gangue stay time in the sorting area,whether the gangue is blocked,gangue confidence,gangue size and other influential factors to determine the optimal target selection sequence,to achieve the rapid sorting of gangue.(5)Development and experimental verification of the robot experimental platform for gangue selection.According to the principle and functional requirements of the robot for picking gangue,an experimental platform for the robot for picking gangue is developed in this paper,which consists of five modules: distributing,transporting,identifying,sorting and receiving module.The platform is used to verify the effectiveness of the proposed algorithm for real-time recognition of coal gangue and quick sorting of coal gangue.When the belt speed is 20mm/s,the identification rate and sorting rate of coal and gangue with particle size of 20-50 mm can reach more than 95%,realizing realtime detection and sorting of coal and gangue.Figure 82 Table 23 Reference 164... |
PDF Full Text Request