Design Of The Recognition And Counting System Of Quasi-circular Granule Images Based On Android Platform

Recognition and counting technology of quasi-circular granule images has been widely used in the fields of agriculture,industry and health care,e.g.measurement of 1000 grain weight of oil seeds,evaluation of fruit yield,calculation of the number of steel bars,counting of the number of cells,etc.However,the traditional machine vision system has many limitations,e.g.high hardware cost,complex building,not convenient mobile operation,etc.These limitations are particularly prominent in fast and real-time counting situations.Therefore,in order to increase practical value of the counting system,a portable Android mobile device was used to study the recognition and counting system of adhesive and overlapped quasi-circular granule images in this paper.Since the granule image is affected by illumination,shadow,etc.,a foreground extraction method based on color difference feature and image enhancement was proposed in this paper.This method not only eliminates the influence of illumination and shadow,but also increases the contrast between granules,and thus it improves the accuracy of foreground pixel extraction.According to the characteristics of granules that their shapes are similar to circular and the large-area and dense adhesion or overlapping exists,a watershed segmentation method based on the distance grayscale image was proposed in this paper.This method can effectively separate the adhesive and overlapping granule regions into individual granules.Among them,some granules are over-segmented due to the existence of holes.In order to improve the accuracy of segmentation,a method for secondary watershed segmentation based on the hole extraction algorithm was proposed in this paper to obtain the segmentation result of granules and their corresponding center points.However,the above segmentation method is based on the binary image of overlapping granules that lose edge information,and thus this method has errors in the segmentation of overlapping granules.To this problem,the overlapping granules were separated from original color images on the basis of segmenting foreground particles and obtaining their center points,and then the rules were defined to screen out the outer contour fragments of the granules from their edge information.Secondly,the contour-seed point association algorithm was used tomatch the contour and center point belonging to the same granule,and then the matched contour points were performed with ellipse fitting,in order to effectively segment the overlapped granules.Finally,the total number of granules in the segmentation results was counted.Based on the Android application programming development technology and the Open CV visual database,the function of each module in the counting system was designed and implemented in this paper,and then the system was carried out the running test and the accuracy and efficiency test of the counting.The test results of the system show that this counting system runs stably.The counting results show that the accuracy of the counting results of adhesive soybean granules is the highest and there is almost no error,while the mean of the counting error of the overlapping soybean granules is 0.23 % and the mean error of the overlapping sand sugar orange is 1 %.Furthermore,the average counting time of the system is 3.36 seconds.This study provides a new idea for exploration of image analysis methods of overlapping quasi-circular granules and portable counting of these granules.