Research On Coal And Gangue Sorting And Identification Algorithm Based On Bio-like Vision

Gangue identification is one of the important applications of gangue sorting,effective identification of gangue can improve coal mine safety,it can avoid serious waste of resources and improve the economic efficiency of the company.Image recognition technology compared to traditional manual sorting and machine sorting,it ensures that gangue sorting is efficient,consumes less energy and avoids causing environmental pollution,so it is extremely important to use gangue image based detection methods to identify the gangue.Due to the advantages of fast recognition,accurate recognition and high sensitivity of bio-vision,it is therefore a proven method for detecting gangue targets,and can quickly and accurately detect the position and size of the target.Therefore,biovision based imitation becomes an important research direction in the field of coal gangue sorting.This thesis builds a coal gangue image recognition dataset based on biovision-like theory,the transmission properties of visual information in the nervous system,It analyses a calibration model for image recognition of coal gangue built on the basis of a camera,the calibration of the camera is completed with the Zhang Zhengyou calibration method under the coordinate transformation relationship in the unified model.The image quality of the gangue will directly affect the recognition effect of the trained network model after the gangue image is captured by the monocular camera.Histogram equalisation and non-linear low-pass filtering are used to enhance the image quality according to the characteristics of the gangue image samples,and the separation and reconstruction of the gangue samples is completed by the erosion and expansion algorithm in morphology to obtain independent targets.Deep learning algorithms can achieve bio-vision recognition during target detection,improved YOLOv3 algorithm for gangue image recognition with different size and shape of gangue in the image.Thus,a Cspdarknet-based gangue target detection model was established,and then the YOLOv3,YOLOv5 and YOLOv3-Cspdarknet algorithms were used to identify and classify the gangue images respectively,and the recognition accuracy of the three algorithms was compared.The algorithm performs image segmentation and position localisation on the identified gangue image.Automatic segmentation using Grab Cut algorithm in combination with YOLOv3,and based on image segmentation and then using the gangue contour centre matching method to localise the gangue,the identification and localisation of targets in gangue sorting is finally completed.The results of the study show that the YOLOv3-Cspdarknet algorithm model and the combination of this algorithm with the Grabcut algorithm can be used to identify and locate coal gangue,based on the principle of biovision and the monocular camera as hardware device in combination with bio-vision principles as the hardware device.The activation function in the original model was changed to Mish and Dark Net-53 adds CSPnet to form Cspdarknet algorithm,it solves the problems of large feature image size,small perceptual field area and inaccurate prediction.Improved Cspdarknet model achieves 98% recognition rate and 47 fps recognition rate,the Grabcut algorithm combined with the Cspdarknet model was used to achieve an error rate of no more than 1.3% when automatically cutting images and localising them.