An intelligent bullet sorting system

Abstract

The present application relates to a kind of intelligent bullet sorting systems, including feeding transmission part, image acquisition part, intelligent identification detection part, communication part, control part, sorting execution part, feeding transmission part includes conveyor belt, tray and motion control system, image acquisition part includes line scanning camera, lens, light source system, image acquisition card, rotary encoder, intelligent identification detection part includes detection algorithm module, detection algorithm module is used to detect all different types of bullets in the entire tray field of view range;Communication part includes database and communication module, and sorting execution part includes manipulator, and different types of bullets are grabbed to corresponding position by controlling manipulator to realize the sorting of bullet.The present application can realize the online, fast, accurate real-time detection sorting of bullet, can automatically create template for newly detected bullet model, and carry out data statistics and analysis.

Description

Technical Field

[0001] This invention relates to an intelligent bullet sorting system, belonging to the field of bullet sorting technology. Background Technology

[0002] In recent years, machine vision inspection technology has begun to play an important role in the inspection of many products. A crucial aspect of military modernization is the recycling and reuse of ammunition. However, due to the diverse types of weapons and ammunition models, current ammunition recycling and sorting methods primarily rely on simple manual visual inspection. This method is labor-intensive, prone to visual fatigue leading to misclassification and inconsistent results; the results are highly dependent on worker experience, making process control and statistical analysis extremely difficult; and it hinders the automated recycling and sorting of ammunition, failing to meet the demands of industrialization and automation. In modern ammunition recycling and reuse, military production units urgently need a highly efficient and intelligent system to replace manual labor.

[0003] Based on this, the present invention is proposed. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by providing an intelligent bullet sorting system, the specific technical solution of which is as follows:

[0005] An intelligent bullet sorting system includes:

[0006] The feeding and conveying section includes a conveyor belt, a tray, and a motion control system. Different types of bullets are placed in the tray, and the movement of the conveyor belt is controlled by the motion control system.

[0007] The image acquisition section includes a line scan camera, lens, light source system, image acquisition card, and rotary encoder. The tray is transported to the image acquisition section via a conveyor belt, and the image acquisition section acquires the original image of the tray containing bullets.

[0008] The intelligent identification and detection section includes a detection algorithm module, which is used to detect all different types of bullets within the entire field of view of the tray;

[0009] The communication section includes a database and a communication module, wherein the database is used to store bullet type, location information and detection results in real time;

[0010] In the control section, bullet type, location information, and detection results are communicated to the control section via a communication module;

[0011] The sorting execution unit includes a robotic arm. The control unit receives the detection results and issues corresponding instructions to the robotic arm. By controlling the robotic arm, different types of bullets are picked up and placed in the corresponding positions to achieve bullet sorting.

[0012] As a further improvement, the detection algorithm module includes the following processes:

[0013] Let the size of the detection be S, the diameter of the bullet be R, the distance from the top of the bullet to the line scan camera be L, the focal length of the line scan camera be f, and the number of pixels be N, then

[0014] The speed of the conveyor belt and the line frequency of the line scan camera are matched according to a matching formula, which is:

[0015]

[0016] L w L is the length of the pallet along the direction of the conveyor belt. h P is the length of the pallet perpendicular to the direction of the conveyor belt's movement. w λ is the number of pixels on one side of the line scan camera, V is the speed of the conveyor belt, t is the time required for the tray to go from entering the line scan camera's field of view to leaving the line scan camera's field of view, R is the radius of the rotary encoder, B is the number of pulses that the rotary encoder can emit in one revolution, and λ is the frequency multiplier that needs to be manually adjusted.

[0017] The algorithm in the detection algorithm module includes the following steps:

[0018] Step 1: Place the empty tray on the conveyor belt and run it. Adjust the parameter of λ so that the image captured by the line scan camera is a complete and unstretched image.

[0019] Step 2: Acquire images Image1 (empty tray) and Image2 (tray containing randomly loaded bullets), and perform image processing on Images Image1 and Image2 respectively.

[0020] First, perform spatial nonlinear filtering on images Image1 and Image2, using Rank... m ×Rank n The size of the matrix A is collected by sliding it from top to bottom and from left to right. p ×Rank q For each pixel in the image, the brightness value of all pixels is obtained; by calling the function fun1, it is converted into a column vector, so that matrix A generates a Rank value for each pixel it passes through. m ·Rank n A vector of dimension Rank p ×Rank q The image will generate Rank p ·Rank q Given such vectors, a Rank can be obtained. m·Rank n ·Rank p ·Rank q The matrix is ​​given; each column of the matrix is ​​substituted into the function fun2 to generate a Rank. p ·Rank q A dimensional vector, then the Rank is obtained through the inverse operation of the function fun1. p ·Rank q The dimensional vector is restored to Rank. p ×Rank q Given a matrix, perform the calculations.

[0021] Image enhancement is performed on images Image1 and Image2.

[0022]

[0023] I represents the pixel component value of the original image. * The component values ​​corresponding to the changed image are denoted by n, where n is the minimum gray value of the bullet.

[0024] An improved fast adaptive image binarization method is introduced to perform image binarization processing:

[0025] G(x,y) is the gray value of the current pixel (x,y), S(x,y) is the sum of the gray values ​​of all pixels to the left and above the current pixel (x,y), x represents the coordinate of the current pixel in the x-direction in the coordinate system, and y represents the coordinate of the current pixel in the y-direction in the coordinate system.

[0026] By iterating through all pixels using S(x,y)=G(x,y)+S(x-1,y)+S(x,y-1)-S(x-1,y-1), and then using these pixels as the center, the average gray value of the rectangle is calculated and compared with the current pixel to perform image segmentation, resulting in the segmented images Image11 and Image21.

[0027] Step 3: Extract the contours of Image11 using Hough circle detection to extract the circular areas where the 6*8 bullets on the tray are placed. Perform erosion and dilation operations on each circular area to obtain the ROI (Region of Interest). i With the top left corner of the tray as the origin, the positions of each circular area are denoted by Cir. i (x,y), i = 1,2,...48, where i is an integer;

[0028] Step 4: Using Cir i Using (x,y) as a template, after performing translation correction on image Image21, the image in the detection region is obtained. For images Model classification j = 1, 2, ..., t, where j is an integer and t is the class; then feature extraction is performed to obtain the initial image of bullet class j.

[0029] right Decomposition yields the scale coefficients. k = 1, 2, 3, 4, 5, for Thresholding is performed based on function F to obtain F(·). Then, all F(·) of this category are reconstructed and weighted to obtain the initial image of bullet j.

[0030] Step 5: After acquiring the initial images of various types of bullets, perform real-time detection and acquisition of bullets on the newly arrived tray. Repeat steps 2 to 4 to obtain the current real-time image I. current The initial image of bullet type j The data of each pixel is the same as the current real-time image I. current Perform function The operation yields the desired information E. j Regarding the expected information E j The similarity is used to make a judgment. A judgment result P(j) of 1 indicates that it is an existing model. For existing models, the coordinate information is returned directly. For new models, a template is automatically created and the user is prompted.

[0031]

[0032] col is the column of the image, and row is the height of the image;

[0033]

[0034] γ is the set similarity threshold, which is set to 0.8.

[0035] A further improvement is made when the maximum diameter of the bullet to be detected is 20mm, then R max The maximum accuracy S required for detection is 20mm. max If the thickness is 0.03mm, then the maximum number of pixels N is... max =666.7, the minimum resolution of a line scan camera is 667*667, the minimum resolution of a line scan camera is 1334*1334, and the minimum resolution of a line scan camera with a narrow edge is 8004.

[0036] In a further improvement, the light source system includes an open backlight located directly above the tray and parallel line light sources located on the left and right sides above the tray in an axisymmetric manner. The two parallel line light sources are arranged facing each other and the angle between the two parallel line light sources and the plane of the tray is 45°. The light source system is cooled by a combination of strobe and water cooling.

[0037] In a further improvement, the image acquisition card is used for image acquisition and transmission; the rotary encoder is used to provide acquisition signals to the line scan camera.

[0038] The beneficial effects of this invention are:

[0039] 1. This invention relates to the application of machine vision in the field of military product inspection. It achieves online, rapid, accurate and real-time detection and sorting of bullets by acquiring images, processing raw images, and extracting and judging features.

[0040] 2. This system can automatically create templates for newly detected bullet types and perform data statistics and analysis. The algorithm in this invention involves extensive matrix operations, making it suitable for parallel acceleration and solving the real-time problem of handling massive amounts of data.

[0041] 3. This invention can replace manual labor in the recycling and sorting of bullets, freeing up manpower, improving efficiency, and enhancing safety.

[0042] 4. It has an intelligent model matching function, which collects, processes and stores images of bullets of different models, and can automatically recall them when the corresponding model is encountered without having to match them again.

[0043] 5. It is robust and can be installed and debugged once the initial setup is completed, without the need for multiple adjustments, regardless of the transmission line or speed. Attached Figure Description

[0044] Figure 1 This is an architecture diagram of the intelligent bullet sorting system described in this invention;

[0045] Figure 2 This is a schematic diagram of the tray described in this invention;

[0046] Figure 3 This is a schematic diagram of the light source system described in this invention;

[0047] Figure 4 A flowchart for the detection algorithm module;

[0048] Figure 5 This is the original image of the bullet;

[0049] Figure 6 This is the outline of the bullet after it has been processed using a global threshold segmentation algorithm.

[0050] Figure 7 This is the bullet's outline image after processing using the Canny algorithm;

[0051] Figure 8 It is a contour image processed using the algorithm of this invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] Example 1

[0054] like Figure 1 As shown, the intelligent bullet sorting system includes:

[0055] The feeding and conveying section includes a conveyor belt, a tray, and a motion control system. Different types of bullets are placed in the tray, and the motion control system controls the movement of the conveyor belt. This feeding and conveying section can place different types of bullets in a 600*800mm container. Figure 2 The tray shown is transported to the image acquisition section via a conveyor belt.

[0056] The image acquisition section includes a line scan camera, lens, light source system, image acquisition card, and rotary encoder. The tray is transported to the image acquisition section via a conveyor belt, and the image acquisition section can acquire high-quality, high-contrast original images of the tray loaded with bullets.

[0057] Since the tray size is 600*800mm and there may be multiple types of bullets on the same tray, a line scan camera was chosen to acquire the image. This line scan camera can avoid the occlusion phenomenon in image imaging caused by the inconsistent height of the bullets, and it can also use the image of a fixed row to be acquired each time to stitch together the images to achieve the acquisition of the whole picture.

[0058] The image acquisition card is used for image acquisition and transmission; the data volume of the line scan camera is too large, and ordinary gigabit networks cannot meet the real-time transmission requirements. The image acquisition card can acquire and transmit data without loss.

[0059] The rotary encoder is used to provide acquisition signals to the line scan camera. Within the same cycle, the number of pulses supplied by the rotary encoder determines the number of 8192*1 images acquired by the line scan camera.

[0060] The light source system includes an open backlight located directly above the tray and parallel line light sources located on the left and right sides above the tray in an axisymmetric manner. The two parallel line light sources are arranged facing each other and the angle between the two parallel line light sources and the plane of the tray is 45°. The light source system is cooled by a combination of strobe and water cooling.

[0061] Because the light source is at an angle to the surface of the illumination tray, and the light source emits parallel rays, the distance from the light source to the nearest and farthest sides is different, resulting in uneven illumination intensity per unit area. This invention, however, uses two parallel light sources positioned opposite each other at a specific angle to ensure uniform illumination intensity in the central area.

[0062] The intelligent identification and detection section includes a detection algorithm module, which is used to detect all different types of bullets within the entire field of view of the tray.

[0063] The communication section includes a database and a communication module, wherein the database is used to store bullet type, location information and detection results in real time;

[0064] In the control section, bullet type, location information, and detection results are communicated to the control section via a communication module;

[0065] The sorting execution unit includes a robotic arm. The control unit receives the detection results and issues corresponding instructions to the robotic arm. By controlling the robotic arm, different types of bullets are picked up and placed in the corresponding positions to achieve bullet sorting.

[0066] Example 2

[0067] The detection algorithm module includes the following processes:

[0068] Let the size of the detection be S, the diameter of the bullet be R, the distance from the top of the bullet to the line scan camera be L, the focal length of the line scan camera be f, and the number of pixels be N, then For example: if the maximum diameter of the bullet to be tested is 20mm, then R max The required accuracy S for the measurement is 20mm. max If it is 0.03mm, then N max =666.7. Based on this precision, the camera's resolution must be at least 667*667. In order to preserve the complete original information, the image needs to fall on more than 2 pixels. Calculated with 2 pixels, the minimum resolution of the camera is 1334*1334. Since the bullets in the tray are arranged in a 6*8 pattern, the minimum resolution of the camera's narrow side is 8004. Therefore, the Hikvision 8K line scan camera is selected, with a resolution of 8192*M (M is the number of stitched rows), which can meet the resolution requirements for bullet detection.

[0069] The accuracy of the detection results largely depends on the quality of the images captured by the camera. Higher image contrast makes it easier to determine the bullet's size and model. Because bullets exhibit metallic reflections, a design was implemented to achieve uniform illumination over a large area of ​​the entire tray, highlighting the bullet's surface features. Figure 3The light source system features an open-aperture backlight to improve overall brightness, and two opposing parallel line light sources on the left and right to reduce interference from reflections. To ensure that the temperature of the light source does not get too high on site, a dual method of strobe and water cooling is used to cool the light source.

[0070] Field verification showed that a row count of 8192*11000 can completely capture the data from a 600*800 tray. A single image from a line scan camera is 8192*1, and stitching together 11000 images results in 8192*11000; 11000 represents the number of signals that need to be acquired.

[0071] The specific implementation steps of the algorithm are as follows:

[0072] Because a line scan camera requires relative motion with the object to capture a complete image, and to prevent image stretching and compression, its speed and line frequency must be matched. The conveyor belt speed and the line scan camera's line frequency are matched according to a formula:

[0073]

[0074] L w L is the length of the pallet along the direction of the conveyor belt. h P is the length of the pallet perpendicular to the direction of the conveyor belt's movement. w V is the number of pixels on one side of the line scanner, t is the speed of the conveyor belt, R is the time required for the tray to move from the line scanner's field of view to its departure from the line scanner's field of view, B is the radius of the rotary encoder, λ is the number of pulses that the rotary encoder can emit in one revolution, and λ is the frequency multiplier that needs to be manually adjusted.

[0075] like Figure 4 As shown, the algorithm in the detection algorithm module includes the following steps:

[0076] Step 1: Place the empty tray on the conveyor belt and run it. Adjust the parameter of λ so that the image acquired by the line scan camera is a complete and unstretched image.

[0077] Step 2: Acquire images Image1 (empty tray) and Image2 (tray containing randomly loaded bullets), and perform image processing on Images Image1 and Image2 respectively.

[0078] First, perform spatial nonlinear filtering on images Image1 and Image2, using Rank... m ×Rank n The size of the matrix A is collected by sliding it from top to bottom and from left to right. p ×Rank qFor each pixel in the image, the brightness value of all pixels is obtained; by calling the function fun1, it is converted into a column vector, so that matrix A generates a Rank value for each pixel it passes through. m ·Rank n A vector of dimension Rank p ×Rank q The image will generate Rank p ·Rank q Given such vectors, a Rank can be obtained. m ·Rank n ·Rank p ·Rank q The matrix is ​​given; each column of the matrix is ​​substituted into the function fun2 to generate a Rank. p ·Rank q A dimensional vector, then the Rank is obtained through the inverse operation of the function fun1. p ·Rank q The dimensional vector is restored to Rank. p ×Rank q The matrix is ​​used to perform the calculation. Where Rank m Rank n These are the rows and columns of the matrix, respectively; Rank p Rank q These are the rows and columns of the matrix, respectively.

[0079] Since the captured bullet images have a narrow grayscale dynamic range and insignificant changes, classifying them as low-contrast images, image enhancement will be performed on images Image1 and Image2 as follows:

[0080]

[0081] I represents the pixel component value of the original image. * The component values ​​corresponding to the changed image are denoted by n, where n is the minimum gray value of the bullet.

[0082] An improved fast adaptive image binarization method is introduced to perform image binarization processing:

[0083] G(x,y) is the gray value of the current pixel (x,y), S(x,y) is the sum of the gray values ​​of all pixels to the left and above the current pixel (x,y), x represents the coordinate of the current pixel in the x-direction in the coordinate system, and y represents the coordinate of the current pixel in the y-direction in the coordinate system.

[0084] By iterating through all pixels using S(x,y)=G(x,y)+S(x-1,y)+S(x,y-1)-S(x-1,y-1), and then using these pixels as the center, the average gray value of the rectangle is calculated and compared with the current pixel to perform image segmentation, resulting in the segmented images Image11 and Image21.

[0085] Step 3: Extract the contours of Image11 using Hough circle detection to extract the circular areas where the 6*8 bullets on the tray are placed. Perform erosion and dilation operations on each circular area to obtain the ROI (Region of Interest). i With the top left corner of the tray as the origin, the positions of each circular area are denoted by Cir. i (x,y), i=1,2,...48, where i is an integer.

[0086] Step 4: Using Cir i Using (x,y) as a template, after performing translation correction on image Image21, the image in the detection region is obtained. For images Model classification j = 1, 2, ..., t, where j is an integer and t is the class; then feature extraction is performed to obtain the initial image of bullet class j.

[0087] right Decomposition yields the scale coefficients. k = 1, 2, 3, 4, 5, for Thresholding is performed based on function F to obtain F(·). Then, all F(·) of this category are reconstructed and weighted to obtain the initial image of bullet type j.

[0088] Step 5: After acquiring the initial images of various types of bullets, perform real-time detection and acquisition of bullets on the newly arrived tray. Repeat steps 2 to 4 to obtain the current real-time image I. current The initial image of bullet type j The data of each pixel is the same as the current real-time image I. current Perform function The operation yields the desired information E. j Regarding the expected information E j The similarity is used to make a judgment. A judgment result P(j) of 1 indicates that it is an existing model. For existing models, the coordinate information is returned directly. For new models, a template is automatically created and the user is prompted.

[0089]

[0090] col is the column of the image, and row is the height of the image;

[0091]

[0092] γ is the set similarity threshold, which is set to 0.8.

[0093] In this invention, a specific algorithm is used to enhance images Image1 and Image2. This is because the acquired bullet images have a narrow grayscale dynamic range and do not change significantly, making them low-contrast images. By setting the minimum grayscale value in the bullet image to 0 and setting the part symmetrical about the maximum grayscale value of 255 to 255, the grayscale value range of the bullet image is expanded, thereby improving the grayscale contrast of the bullet image and facilitating subsequent contour extraction.

[0094] Other image enhancement algorithms, such as:

[0095] 1. Image enhancement based on gamma transform is effective for images with low contrast and high overall brightness (due to camera overexposure).

[0096] 2. Image enhancement based on object log transform can expand the low grayscale part of an image, revealing more details in the low grayscale part, while compressing its high grayscale part, reducing the details in the high grayscale part, thereby achieving the purpose of emphasizing the low grayscale part of the image.

[0097] 3. Image enhancement based on histogram equalization distributes the grayscale image of the original image evenly across the entire grayscale space from a relatively concentrated grayscale range, achieving non-linear stretching of the image and redistributing image pixel values.

[0098] None of these methods can clearly distinguish the grayscale of the bullet image.

[0099] In this invention, steps four and five can improve the bullet classification effect. If, in step four, the image ROI is... i * The method of fitting a circle using least squares to find the largest circle, directly returning coordinate information for existing models and automatically creating a template and prompting the user for new models; however, this only classifies bullets based on their outermost circle, resulting in bullets with the same outer diameter but different inner diameters being grouped into the same category. This invention, through step four, extracts features from multiple dimensions, enabling accurate bullet classification.

[0100] γ is set to a value that has been continuously adjusted based on extensive actual measurements. If γ is set too small, such as 0.5, it will cause some bullet images to be misclassified, for example, a bullet that is actually of type A might be classified as a bullet of type B, and the probability of this happening increases exponentially. If γ is set too large, such as 0.9, some bullet images may not find the established category and will automatically generate a new category, failing to achieve the desired accurate classification.

[0101] In step four, for Decomposition yields the scale coefficients. k = 1, 2, 3, 4, 5, with a maximum value of 5. When k is too small, the extracted features are insufficient. For example, when k = 3, only edge information can be extracted. When k is too large, the computation time is too long due to the exponential growth of the calculation. For example, when k = 7, the detection time is 20 times that when k = 5, close to 10 seconds, which affects the detection performance.

[0102] Figure 5 This is the original image of the bullet. Figure 6 This is the outline of the bullet after being processed using a global threshold segmentation algorithm (first, a segmentation threshold is set based on the grayscale histogram of the image, then the grayscale values ​​of each pixel in the image whose grayscale value is lower than the segmentation threshold are set to 0, and the rest are set to 255). Figure 7 This is the contour image of the bullet after the contour was extracted using the Canny algorithm (first, Gaussian blur is applied to remove noise from the image, then the gradient magnitude and direction of the image are calculated, and finally, non-maximum suppression is used to obtain the edge contour). Figure 8 This is the contour image processed using the algorithm of this invention. Figures 5-8 It is known that conventional threshold segmentation cannot effectively filter out interference points inside an image; conventional edge extraction will lose some contour information; this invention can remove interference points inside the image as much as possible while preserving the contour.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent bullet sorting system, characterized in that, include: The feeding and conveying section includes a conveyor belt, a tray, and a motion control system. Different types of bullets are placed in the tray, and the movement of the conveyor belt is controlled by the motion control system. The image acquisition section includes a line scan camera, lens, light source system, image acquisition card, and rotary encoder. The tray is transported to the image acquisition section via a conveyor belt, and the image acquisition section acquires the original image of the tray containing bullets. The intelligent identification and detection section includes a detection algorithm module, which is used to detect all different types of bullets within the entire field of view of the tray; The communication section includes a database and a communication module, wherein the database is used to store bullet type, location information and detection results in real time; In the control section, bullet type, location information, and detection results are communicated to the control section via a communication module; The sorting execution part includes a robotic arm. The control part receives the detection results and issues corresponding instructions to the robotic arm. By controlling the robotic arm to grab different types of bullets to the corresponding positions, the bullets are sorted. The detection algorithm module includes the following processes: Let the size of the detection be S, the diameter of the bullet be R, the distance from the top of the bullet to the line scan camera be L, the focal length of the line scan camera be f, and the number of pixels be N, then The speed of the conveyor belt and the line frequency of the line scan camera are matched according to a matching formula, which is: L w L is the length of the pallet along the direction of the conveyor belt. h P is the length of the pallet perpendicular to the direction of the conveyor belt's movement. w λ is the number of pixels on one side of the line scanner, V is the speed of the conveyor belt, t is the time required for the tray to go from entering the line scanner's field of view to leaving the line scanner's field of view, R is the radius of the rotary encoder, B is the number of pulses emitted by the rotary encoder in one revolution, and λ is the frequency multiplier that needs to be manually adjusted. The algorithm in the detection algorithm module includes the following steps: Step 1: Place the empty tray on the conveyor belt and run it. Adjust the parameter of λ so that the image captured by the line scan camera is a complete and unstretched image. Step 2: Acquire images Image1 (empty tray) and Image2 (tray containing randomly loaded bullets), and perform image processing on Images Image1 and Image2 respectively. First, perform spatial nonlinear filtering on images Image1 and Image2, using Rank... m ×Rank n The size of the matrix A is collected by sliding it from top to bottom and from left to right. p ×Rank q For each pixel in the image, the brightness value of all pixels is obtained; by calling the function fun1, it is converted into a column vector, so that matrix A generates a Rank value for each pixel it passes through. m ·Rank n A vector of dimension Rank p ×Rank q The image will generate Rank p ·Rank q Given such vectors, a Rank can be obtained. m ·Rank n ●Rank p ●Rank q The matrix is ​​given; each column of the matrix is ​​substituted into the function fun2 to generate a Rank. p ●Rank q A dimensional vector, then the Rank is obtained through the inverse operation of the function fun1. p ●Rank q The dimensional vector is restored to Rank. p ×Rank q Given a matrix, perform the calculations. Image enhancement is performed on images Image1 and Image2. I represents the pixel component value of the original image. * The component values ​​corresponding to the changed image are denoted by n, where n is the minimum gray value of the bullet. An improved fast adaptive image binarization method is introduced to perform image binarization processing: G(x,y) is the gray value of the current pixel (x,y), S(x,y) is the sum of the gray values ​​of all pixels to the left and above the current pixel (x,y), x represents the coordinate of the current pixel in the x-direction in the coordinate system, and y represents the coordinate of the current pixel in the y-direction in the coordinate system. By iterating through all pixels using S(x,y)=G(x,y)+S(x-1,y)+S(x,y-1)-S(x-1,y-1), and then using these pixels as the center, the average gray value of the rectangle is calculated and compared with the current pixel to perform image segmentation, resulting in the segmented images Image11 and Image21. Step 3: Extract the contours of Image11 using Hough circle detection to extract the circular areas where the 6*8 bullets on the tray are placed. Perform erosion and dilation operations on each circular area to obtain the ROI (Region of Interest). i With the top left corner of the tray as the origin, the positions of each circular area are denoted by Cir. i (x,y), i = 1,2,...48, where i is an integer; Step 4: Using Cir i Using (x,y) as a template, after performing translation correction on image Image21, the image in the detection region is obtained. For images Model classification j = 1, 2, ..., t, where j is an integer and t is the class; then feature extraction is performed to obtain the initial image of bullet class j. right Decomposition yields the scale coefficients. k = 1, 2, 3, 4, 5, for Thresholding is performed based on function F to obtain F(·). Then, all F(·) of this category are reconstructed and weighted to obtain the initial image of bullet type j. Step 5: After acquiring the initial images of various types of bullets, perform real-time detection and acquisition of bullets on the newly arrived tray. Repeat steps 2 to 4 to obtain the current real-time image I. current The initial image of bullet type j The data of each pixel is the same as the current real-time image I. current Perform function The operation yields the desired information E. j Regarding the expected information E j The similarity is used to make a judgment. A judgment result P(j) of 1 indicates that it is an existing model. For existing models, the coordinate information is returned directly. For new models, a template is automatically created and the user is prompted. col is the column of the image, and row is the height of the image; γ is the set similarity threshold, which is set to 0.

8.

2. The intelligent bullet sorting system according to claim 1, characterized in that: When the maximum diameter of the bullet to be tested is 20mm, then R max The maximum accuracy S required for detection is 20mm. max If the thickness is 0.03mm, then the maximum number of pixels N is... max =666.7, the minimum resolution of a line scan camera is 667*667, the minimum resolution of a line scan camera is 1334*1334, and the minimum resolution of a line scan camera with a narrow edge is 8004.

3. The intelligent bullet sorting system according to claim 1, characterized in that: The light source system includes an open backlight located directly above the tray and parallel line light sources located on the left and right sides above the tray in an axisymmetric manner. The two parallel line light sources are arranged facing each other and the angle between the two parallel line light sources and the plane of the tray is 45°. The light source system is cooled by a combination of strobe and water cooling.

4. The intelligent bullet sorting system according to claim 1, characterized in that: The image acquisition card is used for image acquisition and transmission; the rotary encoder is used to provide acquisition signals to the line scan camera.

Images