Granularity Estimation of Industrial Products Using Texture Analysis
Texture analysis methods are widely used in various monitoring and measurement tasks in machine vision solutions. Texture analysis tools can be applied also to the estimation of granular sizes of the particle distributions occurring in several kinds of industrial processes. In this article, a texture-based approach for the determination of grain size distributions in the manufacturing processes is presented. This method is based on gray level statistics in the texture image.
1. Introduction
Materials in various industrial processes can be effectively characterized and their visual properties can be measured using image analysis and machine vision methods. In the process industry, significant amounts of information on the process can be acquired using machine vision. This information is utilized in process monitoring and control tasks.
The inspection and measurements of granular products is a fundamental problem in several industrial processes. Traditionally, the characterization of the granular products has been carried out by manual inspection. That is, the grain properties have been inspected by sieving or microscope analysis. This kind of approach is time consuming and hence unable to provide on-line information from the granulation process. For this reason, image analysis methods have been adopted to inspection of grain properties in several industrial processes. Examples of industrial applications in this field can be found in pharmaceutical industry, pulp and paper industry, chemical industry, and food industry.
Image analysis applications are also widely being adopted in other industrial areas related to crystallization process, for example. In these kinds of processes, the average size and size distribution of the product play a constitutive role. The properties of the crystal population are important as well for the end use functionality as for the ease of downstream processing.
When image analysis is utilized in the analysis of grain properties of crystalline products, one needs to select between two kinds of approaches. The traditional approach is to extract the grains from the image by using some image segmentation method, thresholding for instance. The grain properties, such as size or shape, are then measured from the segmented image. However, there are several difficulties with this kind of approach. Firstly, the image segmentation algorithms are often sensitive to illumination changes in the process. In addition, the color (or gray level) distributions of the grains are not always homogenous. On the contrary, the colors of the grains to be extracted from the images may vary significantly. Secondly, the segmentation causes computational load that may be critical in the case of on-line analysis and inspection applications. An alternative for the extraction of the grains from the image background is the employment of texture analysis in the granularity measurements. It has been found that texture analysis methods can be applied to the analysis and inspection of granular products. Using texture analysis tools, it is often advantageous to inspect the particle populations as larger surfaces, not by extracting single grains from the image.
2. Texture analysis
Texture analysis is common in various industrial machine vision applications. Texture analysis is used to e.g. estimate different properties of surfaces. Typical industrial applications of texture analysis include the inspection of different surface materials such as paper, metal or textiles. In these applications, the classification and recognition of different surfaces is based on texture properties, for example roughness, granular size or directionality of the texture.
Numerous techniques have been proposed for texture description [2]. In this study, we use statistical texture analysis methods. Statistical techniques are based on the description of the spatial organization of the image gray levels. On the basis of the gray level distribution, it is possible to calculate several types of simple statistical features. Examples of these measures are gray level co-occurrence matrix [1] and gray level difference method [3]. These methods are based on gray level statistics that use pixel pairs at certain distance d.
When the grain size distribution is considered, the use of fixed spatial distance is not necessarily the best choice. On the other hand, the gray level statistics can be calculated by using varying spatial distances. In our approach, maximum difference histogram (MDH), the grain size distribution is presented in the form of histogram. Hence, the MDH presents an estimate of normalized grain size distribution of the image. The calculation of MDH is based on gray level statistics and it can be employed in the estimation of the grain size distribution. If only the average of the grain sizes in the population is interesting, it is possible to be determined based on the histogram. This can be done by determining the center of gravity (CoG) of the histogram. Figure 1 presents an example image of grain population obtained from industrial carbohydrate crystallization process and the MDH histogram calculated based on it. The example image was provided by Danisco Texturants & Sweeteners. In the MDH presented in figure 1, relative proportion of the grain sizes in the carbohydrate image can be seen.
References
|
