On-line Identification And Particle Size Estimation Of Solution Crystal Based On Image Processing

The crystallization technology covers industries such as chemical,food,microelectronics and pharmaceuticals.In order to control the crystallization process effectively,the method based on image processing technology is applied to on-line detection of crystallization process.For the real-time monitoring of crystallization process using image recognition method,due to the interferences of external environment,the image is of noisy,and the complete information of the target area cannot be extracted.With the crystallization process processing,the volume of crystals becomes larger and the number increases,crystals’ overlap gets more.In addition,since the image acquired by a single camera only includes the two-dimensional information of crystal particles,part of the shape informtion is lost,which results in a large error of the obtained particle size distribution.Therefore,based on the existing image processing technologies,combined with actual crystallization processes,this paper improves algorithms to extract the target area of crystals,segments the overlapping area,modifies the length characteristics,and obtains the results of particle size distribution.In order to accurately extract the particle target region,this paper proposes a multi-operator fusion algorithm,which superimposes the detection results of different scales of different operators,obtains more information of the target regon,then deletes and fiters this region.In order to increase the accuracy of the statistical results of particle size distribution,for the problem of particle overlap,we first propose a method to distinguish particle adhesion and particle occlusion based on the variance of gray distribution,and then use an improved concave point segmentation method to segment the overlap region.The concave point detection method locates the concave points based on contour and neighborhood information.Then we use the improved matching rules to match the concave points,and use the improved Bresenham algorithm to permanently segment the overlapping regions.For the crystallization process with crystal transformation,the particle size distribution of different crystal particles needs to be statistically separated.Therefore,we first propose a shape description operator to characterize the shape,use the principal component analysis method to select the classification features,and then use the single-layer BP neural network for classification.In order to increase the accuracy of statistical results,for the large measurement error of the length characteristics of needle-like particles,we first propose to use gray-scale variance product to represent the clearness,and then establish the relationship between clearnessharpness and distance.Based on the function,the depth information of the needle-like crystal particles is obtained.Finally,the actual length of the needle-shaped crystal particles is calculated according to the triangular relationship model,and the data correction is realized.Finally,the statistical results of the classified particle size distribution are obtained based on the image processing technologies,and the histogram of the particle number distribution corresponding to the actual length of the crystal particle is obtained by unit pixel calibration.The statistical results of particle size distribution obtained in this paper provide data support for crystallization control.