Research And Implementation Of Cocoon Counting Method Based On Image Processing

The group value of silkworm cocoon is mainly reflected in the statistics of cocoon number.The statistical results have important reference significance in fresh cocoon purchasing,silk reeling and quality inspection.Traditional cocoon counting method is done manually.Its shortcomings are low efficiency,high labor cost and difficult to ensure the accuracy of counting.The silkworm cocoon counting method can be divided into dynamic and static types according to the motion state at the time of detection.The application of the former is: in the workshop production line,cocoons follow the conveyor belt for continuous movement,which belongs to a flow counting.While the latter application is: cocoons are statically distributed on the experimental desktop,and only counting the number of cocoon samples from current batch.At present,some scholars have made theoretical research on automatic counting of silkworm cocoon,and put forward corresponding image recognition methods,but there are still two shortcomings: 1)Less research on the flow counting of cocoons;2)Counting static cocoons,the existing graphic recognition method is still realized by a computer,and the corresponding counting system has defects of high cost,large volume,insufficient system stability and poor real-time effect.In order to solve the first problem mentioned above,a flow counting system for cocoon video image is designed in this paper,which is mainly composed of cocoon adhesion segmentation and particle tracking between adjacent frame images,and is verified and realized by computer.In the process of cocoon adhesion segmentation,an innovative adhesion segmentation method based on pit directional erosion is proposed.Firstly,the pit search method which is based on square template is used to detect and locate the pits.Then,the erosion direction is calculated and and the traditional erosion algorithm is improved by combining with a variety of structural elements.Finally,the adhesion cocoon is successfully separated according to the corrosion path.The method of image segmentation based on directional erosion in concave points can effectively solve the problem that the existing ‘single pit’ can not be matched,and improve the segmentation accuracy.In order to solve the second shortcoming mentioned above,a real-time counting system for cocoon video image is designed in this paper,which mainly consists of image preprocessing,morphological erosion,contour extraction,connected component labeling and so on.It is realized by FPGA hardware technology.The system utilizes the parallelism of circuit structure,so that FPGA can continuously output the processing result of each frame to achieve real-time video counting effect.In specific operation steps,the main research difficulty is the hardware implementation method of connected component labeling.The solution is to use a pixel-by-pixel scanning idea to create a four-connected scan window,and obtain the current pixel label according to the different combinations of the upper neighborhood and the left neighborhood label.Aiming at the problem of how to update the equivalence table,contour perimeter,bounding box and label X coordinate in time,an information searching mechanism based on clock frequency multiplier is designed to help complete the identification of connected blocks in cocoon image.