Research On Particle Size Measurement Method Based On Microscopic Video Image Processin

The rapid measurement of particle size and distribution plays an important role in mineral resource development,pharmaceutical industry,environmental engineering and other fields.Compared with other measurements,the microscopic imaging measurement is more accurate and can obtain the absolute particle size of the particles.At present,the traditional microscopic imaging measurement takes a single frame of particles for image processing and particle size measurement,which cannot meet the real-time measurement requirements.Based on this,this paper innovatively captures particle video under the microscopy system and performs real-time processing of video images.Aiming at the phenomenon of motion blur and repeated statistics in the measurement process,the method of motion blur recovery which bases on the improved generative adversarial network and the method of particle tracking based on improving SORT are proposed respectively.Firstly,water droplets,spherical polystyrene and quartz particles are chosen as representative particles for research.The above-mentioned particles are photographed under a microscope system.Motion blur is unavoidable in particle videos due to the moving sampling disk as it passes under the microscope system.Aiming at the above phenomena,this paper innovatively proposes a motion blur recovery method based on an improved generative confrontation network.The specific improvements are as follows: a combined effective channel attention mechanism is proposed,and a deep separable convolution is used to reconstruct the feature extraction network,and embedded in the network The attention mechanism finally constructs an improved feature pyramid structure,which provides the basis for subsequent particle identification and analysis.Secondly,according to the different image characteristics of the three kinds of particles,different preprocessing and edge extraction algorithms are used respectively.Aiming at the problem of weak edge and multiple edges in the process of quartz grain edge extraction,an improved Canny edge extraction algorithm is proposed.Then,a particle tracking method based on improved SORT is innovatively proposed in order to meet the requirements of particle repetition statistics and real-time measurement.While satisfying the real-time nature of particle measurement,it solves the problem that the same particle is repeatedly counted in consecutive frames or missed or wrongly detected due to the same particle appearing in different positions in consecutive multiple frames of pictures in video shooting.The specific improvements are as follows: on the basis of the correlation matching of the motion information of the target particles,the cascade matching of the appearance information is added;at the same time,the secondary matching operation is introduced to reduce the phenomenon of missing detection of the model.Finally,the traditional microscopic imaging measurement is chosen as the reference calibration experiment,in which the total errors of the two groups of water droplets particle experimental results were 5.7% and 3.2%,and the total errors of the two groups of quartz particle experimental results were 5.6% and 7.1% respectively,indicating that the statistical error of particle size of the proposed method is small.In addition,the three kinds of particles were repeatedly photographed,and the measurement results remained stable in repeated experiments.