Catapult kinetic energy measurement method and device, electronic equipment and storage medium

Abstract

The invention provides a catapult kinetic energy measurement method and device, electronic equipment and a storage medium, and the method comprises the steps: inputting a print image generated by the impact of a catapult launched by a catapult and the bullet correlation information of the catapult into an improved target detection model, obtaining an imprint prediction frame and an imprint prediction category in an imprint image output by the improved target detection model; after the target area image in the print prediction frame is optimized, the actual print area is calculated according to the optimized target area image; calculating the kinetic energy of the catapult according to the mass of the catapult and the flight speed after launching, and calculating the unit area kinetic energy of the catapult according to the kinetic energy of the catapult and the actual print area; the improved target detection model is obtained through training by taking an imprint image sample and corresponding bullet correlation information as samples and taking an imprint category labeling result and an imprint frame labeling result in the imprint image sample as labels. According to the invention, rapid and accurate measurement of kinetic energy measurement is realized.

Description

Methods, devices, electronic equipment and storage media for measuring the kinetic energy of catapults Technical Field

[0001] This invention relates to the field of catapult testing technology, and in particular to a method, apparatus, electronic device and storage medium for measuring the kinetic energy of a catapult. Background Technology

[0002] To ensure children's health and safety, toy safety standards limit the kinetic energy of projectile toys. Testing of projectile toys is conducted according to these standards. Currently, the field of kinetic energy testing for projectile toys faces the following core problems: 1. Traditional manual testing is extremely inefficient and cannot meet the needs of large-scale testing. The traditional testing process relies entirely on manual operation: first, the maximum flight speed of the projectile is measured using a velocimeter; then, paint is applied to the impact surface of the projectile; after hitting graph paper, an imprint is formed; the tester manually counts the number of colored squares to estimate the impact area, and finally calculates the kinetic energy and the kinetic energy per unit area. A single imprint can contain dozens to hundreds of squares, and manual counting is time-consuming and laborious, with a single sample testing time of approximately 30 minutes.

[0003] 2. Manual measurement has poor accuracy and is prone to judgment bias. When manually counting squares, visual fatigue and blurred square boundaries (such as paint smudging) can easily lead to counting errors. For concentric ring marks and edge fracture marks generated by suction cup-shaped bullets, manual measurement cannot accurately delineate the effective area, resulting in a large relative error in area estimation. This leads to deviations in the calculation of kinetic energy and kinetic energy per unit area, affecting the accuracy of the conformity judgment results. This may result in unqualified products entering the market or qualified products being misjudged.

[0004] 3. Existing technologies have significant shortcomings, with insufficient adaptability and automation. Current technologies can only handle regular circular imprints relatively well, calculating the area of ​​a circle using simple edge detection algorithms, but the accuracy rate for recognizing concentric ring-shaped imprints is less than 60%. Furthermore, a complete automated process of "image acquisition, imprint recognition, area calculation, and kinetic energy determination" has not been established; manual input of projectile velocity and mass data and manual screening of qualified imprints are still required, thus failing to completely eliminate reliance on manual labor.

[0005] With the increasing demand for inspection, traditional manual inspection can no longer meet the efficiency requirements for detecting large batches of samples. Existing machine vision inspection solutions still have significant technical shortcomings. For example, they are only suitable for recognizing regularly shaped imprints, and their accuracy in recognizing irregular, qualified imprints is low. Image preprocessing and contour extraction algorithms are not optimized for the special texture features of imprints, easily leading to contour extraction errors and affecting the accuracy of area calculation. Furthermore, general target detection models have not been improved to suit the specific characteristics of projectile toy imprint detection scenarios, resulting in insufficient robustness for detecting small-sized, multi-shaped imprints and making it difficult to meet practical inspection needs.

[0006] Therefore, developing kinetic energy measurement technology for projectile toys with full-process automation, multi-scenario adaptability, and high-precision measurement capabilities has become a key requirement for improving the level of toy safety testing. Summary of the Invention

[0007] This invention provides a method, device, electronic device, and storage medium for measuring the kinetic energy of a catapult, which solves the problems of traditional catapult toy kinetic energy detection relying on manual labor, low efficiency, and poor accuracy, as well as the shortcomings of existing machine vision solutions with insufficient adaptability and low degree of automation, and realizes automatic measurement of catapult toy kinetic energy based on improved machine vision technology.

[0008] This invention provides a method for measuring the kinetic energy of a catapult, comprising: inputting an image of the imprint generated by the impact of a projectile launched by the catapult and the warhead association information of the projectile into an improved target detection model to obtain a predicted bounding box and a predicted category of the imprint in the image output by the improved target detection model; optimizing the target region image within the predicted imprint and calculating the actual area of ​​the imprint based on the optimized target region image; calculating the kinetic energy of the projectile based on its mass and flight velocity after launch, and calculating the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual area of ​​the imprint; the improved target detection model is obtained by training the imprint image samples and the corresponding warhead association information as samples, and using the imprint category labeling results and imprint bounding box labeling results in the imprint image samples as labels.

[0009] According to the catapult kinetic energy measurement method provided by the present invention, the improved target detection model is based on the YOLOv12 model, and a multi-module collaborative attention mechanism combination structure is embedded in the backbone network of the YOLOv12 model. The attention mechanism combination structure includes an adaptive feature enhancement module, a lightweight feature weighting module, and a multi-dimensional feature enhancement module.

[0010] According to the present invention, a method for measuring the kinetic energy of a catapult includes a backbone network comprising a first combined structure of a first C3K2 module, an adaptive feature enhancement module, and a lightweight feature weighting module, and a second combined structure of an adaptive feature enhancement module and a multi-dimensional feature enhancement module; the first C3K2 module, the first combined structure, and the second combined structure are connected sequentially; the neck network of the YOLOv12 model includes a third combined structure, a fourth combined structure, and a fifth combined structure of a second C3K2 module, an adaptive feature enhancement module, and a lightweight feature weighting module; the output of the second combined structure is upsampled and then concatenated with the output of the first combined structure, and the concatenation result is... The third combined structure is input; the output of the third combined structure is upsampled and concatenated with the output of the first C3K2 module, and the concatenation result is input into the fourth combined structure. The output of the fourth combined structure is input into the first detection module in the head network of the YOLOv12 model; the output of the fourth combined structure is concatenated with the output of the third combined structure and then input into the fifth combined structure. The output of the fifth combined structure is input into the second detection module in the head network; the output of the fifth combined structure is concatenated with the output of the second combined structure and then input into the second C3K2 module. The output of the second C3K2 module is input into the third detection module in the head network.

[0011] According to a catapult kinetic energy measurement method provided by the present invention, before inputting the imprint image generated by the impact of the catapult-launched projectile and the projectile's warhead association information into an improved target detection model, the method further includes: inputting imprint image samples and corresponding projectile association information into the improved target detection model to obtain the imprint prediction bounding boxes and imprint prediction categories in the imprint image samples output by the improved target detection model; calculating the classification loss between the imprint prediction categories in the imprint image samples and the corresponding imprint category annotation results, and the bounding box regression loss between the imprint prediction bounding boxes in the imprint image samples and the corresponding imprint bounding box annotation results; obtaining a target loss function based on the component loss and the bounding box regression loss; and updating the parameters of the improved target detection model through backpropagation based on the target loss function until the change value of the target loss function satisfies a preset convergence condition or the number of training iterations reaches a preset threshold.

[0012] According to a catapult kinetic energy measurement method provided by the present invention, the target region image within the imprint prediction box is optimized, including: converting the target region image into a grayscale image, and then using a smoothing algorithm to suppress noise to reduce interference noise introduced during image acquisition; using an adaptive thresholding strategy to binarize the target region image to distinguish the imprint region from the background region and highlight the imprint contour features; removing isolated noise points in the target region image through morphological opening operations, and combining morphological closing operations to eliminate small holes inside the imprint in the target region image, thereby optimizing the imprint boundary morphology.

[0013] According to the present invention, a method for measuring the kinetic energy of a catapult calculates the actual area of ​​the imprint based on an optimized target region image, comprising: detecting the imprint contour in the optimized target region image based on the Suzuki85 boundary tracking algorithm; correcting local abnormal protrusions or depressions in the imprint contour using a contour optimization algorithm to obtain a regular imprint contour; counting the number of pixels within the regular imprint contour; determining the conversion relationship between pixels and actual length based on the resolution parameters of the imprint image acquisition device; and calculating the actual area of ​​the imprint.

[0014] According to a catapult kinetic energy measurement method provided by the present invention, after inputting the imprint image generated by the impact of a projectile launched by the catapult and the projectile's associated information into an improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image output by the improved target detection model, the method further includes: if the proportion of missing imprint edges within the imprint prediction box exceeds a preset missing threshold, the imprint is deemed unqualified, and the imprint prediction box and the corresponding imprint prediction category are removed; if the imprint prediction category is a circular imprint, the ellipticity of the circular imprint is calculated; if the imprint prediction category is a concentric ring imprint, the interlayer spacing of the concentric ring imprint is calculated; if the ellipticity or the interlayer spacing exceeds a preset deviation threshold for the corresponding imprint prediction category, the imprint is deemed unqualified, and the imprint prediction box and the corresponding imprint prediction category are removed.

[0015] This invention also provides a catapult kinetic energy measurement device, comprising: a trajectory prediction module, used to input an image of the trajectory generated by the impact of a projectile launched by the catapult and the projectile's warhead association information into an improved target detection model to obtain a trajectory prediction box and a trajectory prediction category in the trajectory image output by the improved target detection model; an area calculation module, used to optimize the target region image within the trajectory prediction box and calculate the actual area of ​​the trajectory based on the optimized target region image; and a kinetic energy calculation module, used to calculate the kinetic energy of the projectile based on its mass and flight speed after launch, and to calculate the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual area of ​​the trajectory; the improved target detection model is obtained by training with trajectory image samples as samples and the trajectory category labeling results and trajectory box labeling results in the trajectory image samples as labels.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the catapult kinetic energy measurement method as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the catapult kinetic energy measurement method as described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the catapult kinetic energy measurement method as described above.

[0019] The catapult kinetic energy measurement method, device, electronic device and storage medium provided by the present invention have the following beneficial effects compared with the prior art: (1) Strong adaptability of multi-morphological imprints By embedding a multi-module collaborative attention mechanism combination structure in the YOLO series basic model, and dynamically adjusting the feature extraction strategy in combination with the warhead association information, it can accurately identify qualified imprints of different shapes (such as circles, concentric rings, etc.), solve the problem of low accuracy of the prior art in identifying irregular imprints, and significantly improve the model's adaptability to complex detection scenarios.

[0020] (2) High measurement accuracy: Based on the characteristics of imprint images, the image preprocessing process is optimized (adaptive binarization, targeted morphological operations), and combined with boundary tracking and contour optimization algorithms to effectively eliminate noise interference and contour anomalies. Then, the conversion relationship is calibrated through the resolution of the image acquisition device to ensure the accuracy of imprint area calculation. At the same time, unqualified imprints are eliminated through a multi-dimensional judgment mechanism to avoid the influence of invalid data on the calculation results, and further ensure the reliability of kinetic energy measurement.

[0021] (3) High degree of automation throughout the process: Establish data interaction interface between quality measurement equipment, speed measurement equipment, image acquisition equipment and data processing system to realize automatic flow of detection data without manual intervention in data entry, imprint screening and other links; at the same time, it supports automatic generation and output of test reports, which greatly shortens the processing time of a single sample, significantly improves detection efficiency, and meets the detection needs of large batches of samples.

[0022] (4) Good compatibility of testing standards The entire testing process, including the setup of the experimental environment, parameter calculation, and multiple steps of qualification judgment, is designed based on the safety testing standards for projectile toys. It can be directly connected to the existing regulatory system. The standardized test report generated can meet the regulatory needs of different scenarios such as customs inspection and enterprise quality inspection, providing strong technical support for toy safety supervision. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 is one of the flowcharts of the catapult kinetic energy measurement method provided by the present invention; Figure 2 is one of the flowcharts of the catapult kinetic energy measurement method provided by the present invention; Figure 3 is a schematic diagram of the identification results of three types of imprints (standard printed circles, circular imprints, and concentric ring imprints) in the catapult kinetic energy measurement method provided by the present invention; Figure 4 is a schematic diagram of the functional interface of the measurement software in the catapult kinetic energy measurement method provided by the present invention; Figure 5 is a schematic diagram of the structure of the improved target detection model ASC-YOLOv12 in the catapult kinetic energy measurement method provided by the present invention; Figure 6 is a schematic diagram of the contour extraction and area calculation visualization of circular imprint samples in the catapult kinetic energy measurement method provided by the present invention; Figure 7 is a schematic diagram of the structure of the catapult kinetic energy measurement device provided by the present invention; Figure 8 is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] The following describes a catapult kinetic energy measurement method according to the present invention, with reference to Figure 1, including: Step 101, inputting the imprint image generated by the impact of a catapult-launched projectile and the projectile's warhead association information into an improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image output by the improved target detection model; Step 102, optimizing the target region image within the imprint prediction box, and calculating the actual area of ​​the imprint based on the optimized target region image; Step 103, calculating the kinetic energy of the catapult based on its mass and flight speed after launch, and calculating the kinetic energy per unit area of ​​the catapult based on its kinetic energy and the actual area of ​​the imprint; The improved target detection model is obtained by training the imprint image samples and the corresponding projectile association information as samples, and using the imprint category labeling results and imprint box labeling results in the imprint image samples as labels.

[0027] The catapult can be a projectile toy or other device requiring kinetic energy measurement. As shown in Figure 2, in response to the catapult's kinetic energy measurement command, the system acquires the impact image of the catapult and the projectile's associated information. The projectile's associated information can include the projectile's material and shape.

[0028] First, set up the experimental environment. Construct a testing environment including a shooting test stand, a launching bracket, and an impact receiving mechanism, ensuring that the relative positions of the catapult and the impact receiving mechanism meet the testing requirements of the relevant standards. Vertically fix the catapult to the receiving mechanism covered with white paper, and adjust the launching bracket to ensure the projectile is a certain distance from the test board, such as 300 mm, ensuring consistent launch conditions for each test.

[0029] Then, the imprint is recorded. A marking medium capable of forming an identifiable imprint is placed on the projectile's tip. The projectile is triggered to impact a receiving mechanism, forming an imprint. The projectile can be circular or suction cup-shaped; this embodiment is not limited to this. For example, red pigment is evenly applied to the surface of the projectile, and the projectile is triggered to impact white paper, forming a circular or concentric ring imprint. If the imprint deforms, overlaps, or has missing edges during the impact, it is marked as an invalid imprint, and the above recording process is repeated.

[0030] Next, image acquisition is performed. Image acquisition equipment is used to obtain images of the imprints on the impact receiving mechanism. A flatbed scanner can be used to acquire the imprint images, with the scanning resolution set to 300 dpi to ensure that the images are free of perspective distortion, have uniform illumination, and that the texture details of the imprints (such as the interlayer boundaries of concentric rings and the edges of circular imprints) are clearly identifiable. At the same time, the projectile type (circular or suction cup-shaped) is recorded as projectile-related information, and the mass data and flight velocity data of the projectile are acquired.

[0031] The steps for obtaining the projectile's flight speed data include: measuring the projectile's maximum flight speed using a speed measuring device that meets the detection accuracy requirements; establishing a data interaction channel between the speed measuring device and the data processing system to achieve automatic acquisition of flight speed data.

[0032] Establish a data interaction interface to enable automatic data transfer between quality measurement equipment, speed measurement equipment, image acquisition equipment, and data processing system, thereby shortening the overall processing time for a single sample. Respond to test report generation commands to automatically generate standardized test reports containing test parameters, imprint images, calculation results, and pass / fail conclusions, supporting report preview, output, and storage functions.

[0033] An improved target detection model is constructed and trained. This embodiment does not limit the type of improved target detection model. The imprint image (circular or suction cup shaped) is input into the trained improved target detection model. The feature extraction strategy is dynamically adjusted in combination with the warhead association information (such as strengthening the corresponding feature channels for imprints of different warhead materials). Qualified imprints of different imprint categories (such as circular and concentric rings) are identified and the target region is located. The identification results are shown in Figure 3.

[0034] For example, an improved object detection model based on the YOLO series of basic models can be constructed, embedding a multi-module collaborative attention mechanism combination structure, configuring model training parameters (such as training batches, number of iterations, etc.) and iteratively training the model, and saving the model parameters after training is completed after the model meets the convergence condition.

[0035] The image of the target area after localization is optimized, including preprocessing (grayscale conversion, noise suppression, binarization) and morphological optimization (opening operation, closing operation). Boundary tracking algorithm and contour optimization algorithm are used to extract and correct the imprint contour. The actual area of ​​the imprint is calculated by calibrating the conversion relationship with the resolution parameters of the image acquisition device.

[0036] By acquiring the mass and velocity data of the projectile, and combining this data with the actual area of ​​the imprint, the kinetic energy of the projectile toy and the kinetic energy per unit contact area are obtained using the kinetic energy calculation formula and the kinetic energy per unit area calculation formula, thus completing the kinetic energy measurement. The measured kinetic energy and kinetic energy per unit area can then be judged as qualified according to testing standards.

[0037] In response to the test report generation command, the desktop measurement software developed based on PyQt5 automatically generates a standardized report. The functional interface of the measurement software is shown in Figure 4. Specific functions include: test parameters: projectile type, mass, flight speed, and scanning resolution; image data: original imprint image and model recognition result image; calculation results: imprint area, kinetic energy, and kinetic energy per unit area; pass / fail conclusion: the judgment result according to the GB 6675.2-2014 standard; it supports report preview, printing, and PDF export, and the total processing time from image acquisition to report generation for a single sample is ≤1 second.

[0038] This embodiment addresses the kinetic energy detection requirements of catapults. Taking catapults with different projectiles as the detection objects, it employs an improved target detection model to accurately identify different types of projectiles based on the imprint images generated by the projectiles and the projectile association information. Combined with image optimization processing technology, it ensures the accuracy of imprint area measurement, thereby achieving rapid and accurate measurement of catapult kinetic energy. The fully automated design significantly improves detection efficiency and solves the problems of time-consuming and error-prone traditional manual detection. It can meet the needs of large-volume detection scenarios such as customs on-site inspection and enterprise quality inspection.

[0039] Based on the above embodiments, the improved target detection model in this embodiment uses the YOLOv12 model as the base model. A multi-module collaborative attention mechanism combination structure is embedded in the backbone network of the YOLOv12 model. The attention mechanism combination structure includes an adaptive feature enhancement module, a lightweight feature weighting module, and a multi-dimensional feature enhancement module to adapt to the recognition needs of different morphological imprints.

[0040] This embodiment improves the target detection model by using the YOLOv12 model as the basic framework, considering the balance between detection accuracy and computational complexity, and employing versions N, S, and M as baseline models. An attention module is added to the backbone network, and a lightweight module is replaced to construct the A2C2f-SimAM-CBAM combined attention module. Specifically: the A2C2f module dynamically adjusts the feature weights of imprints at different scales through adaptive channel and spatial attention mechanisms, enhancing the network's ability to capture targets at multiple scales; the SimAM module uses simple nonlinear operations to achieve adaptive feature weighting without introducing additional parameters, ensuring a lightweight model that meets the requirements for rapid detection; and the CBAM module combines channel attention (highlighting key feature channels such as imprint edges and textures) with spatial attention (strengthening imprint region features) to solve the interlayer interference problem of concentric ring imprints on suction cup-shaped warheads.

[0041] Meanwhile, the original multi-scale feature fusion mechanism and anchor box optimization strategy of YOLOv12 are retained, and the feature extraction capability is enhanced only for the imprint detection scenario.

[0042] Based on the above embodiments, as shown in Figure 5, the backbone network in this embodiment includes a first combined structure of a first C3K2 module, an adaptive feature enhancement module, and a lightweight feature weighting module, and a second combined structure of an adaptive feature enhancement module and a multi-dimensional feature enhancement module; the first C3K2 module, the first combined structure, and the second combined structure are connected sequentially; the neck network of the YOLOv12 model includes a third combined structure, a fourth combined structure, and a fifth combined structure of a second C3K2 module, an adaptive feature enhancement module, and a lightweight feature weighting module; the output of the second combined structure is upsampled and then concatenated with the output of the first combined structure, and the concatenation structure is... The output of the third combined structure is upsampled and concatenated with the output of the first C3K2 module. The concatenation result is input into the fourth combined structure, and the output of the fourth combined structure is input into the first detection module in the head network of the YOLOv12 model. The output of the fourth combined structure is concatenated with the output of the third combined structure and then input into the fifth combined structure. The output of the fifth combined structure is input into the second detection module in the head network. The output of the fifth combined structure is concatenated with the output of the second combined structure and then input into the second C3K2 module. The output of the second C3K2 module is input into the third detection module in the head network.

[0043] Based on the above embodiments, this embodiment further includes, before inputting the imprint image generated by the impact of the projectile launched by the catapult and the projectile's warhead association information into the improved target detection model: inputting the imprint image sample and the corresponding warhead association information into the improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image sample output by the improved target detection model; calculating the classification loss between the imprint prediction category in the imprint image sample and the corresponding imprint category labeling result, and the bounding box regression loss between the imprint prediction box of the imprint image sample and the corresponding imprint box labeling result; obtaining the target loss function based on the component loss and the bounding box regression loss; updating the parameters of the improved target detection model through backpropagation based on the target loss function until the change value of the target loss function meets the preset convergence condition or the number of training iterations reaches the preset threshold.

[0044] The steps for training the improved target detection model include: constructing an imprint sample image dataset, collecting sample images containing qualified imprints of different shapes, labeling the qualified imprint categories and corresponding image identification information, and dividing the dataset into training, validation, and test sets according to a preset ratio. For example, 1000 sets of valid imprint samples are collected, including 500 sets of circular imprints (from circular projectiles) and 500 sets of concentric ring imprints (from suction cup-shaped projectiles). The LabelImg annotation tool is used to label the imprint categories ("circular" and "concentric ring") and bounding boxes. The dataset is then divided into a training set (800 sets), a validation set (100 sets), and a test set (100 sets) in an 8:1:1 ratio. Data enhancement processing such as sharpening and blurring is performed on the training set samples to improve the model's generalization ability.

[0045] Configure model training parameters, including batch size, number of training iterations, initial learning rate, momentum coefficient, and weight decay coefficient. For example, the batch size is 8, the number of training epochs is 100, the initial learning rate is 0.01, the momentum is 0.937, the L2 weight decay is 0.0005, and the learning rate strategy is warmup.

[0046] Model training and validation involve iteratively training the improved YOLOv12 model on the training set. In each iteration, a target loss function, including classification loss and bounding box regression loss, is calculated based on the error between the model's output imprinted bounding boxes, class results, and sample annotation results. Model parameters are updated via backpropagation based on the target loss function until the change in the target loss function meets a preset convergence condition or the number of training iterations reaches a preset threshold. At this point, the model parameters are saved, and training is complete. For example, training stops and model parameters are saved when the change in the target loss function is less than 0.001 for five consecutive iterations or when the number of training iterations reaches 100. After each training iteration, model performance (precision, recall, mAP@0.5, mAP@0.5:0.95) is evaluated using a validation set.

[0047] Based on the above embodiments, this embodiment optimizes the target region image within the imprint prediction box, including: converting the target region image into a grayscale image and then using a smoothing algorithm to suppress noise, thereby reducing interference noise introduced during image acquisition; using an adaptive thresholding strategy to binarize the target region image to distinguish the imprint region from the background region and highlight the imprint contour features; removing isolated noise points in the target region image through morphological opening operations and combining morphological closing operations to eliminate small holes inside the imprint in the target region image, thereby optimizing the imprint boundary morphology.

[0048] The target area image is optimized through the following steps: Grayscale conversion: The color image of the target area is converted to a grayscale image to eliminate color channel interference; Gaussian blurring: A 3×3 Gaussian kernel is used for smoothing to reduce scanner noise and pigment particle interference; Binarization: An adaptive thresholding strategy is used to set pixels with a grayscale value > 240 to white (background) and the rest to black (imprint) to highlight the imprint outline; Morphological optimization: Isolated noise points are removed by morphological opening operations of 3×3 structuring elements, and small holes inside the imprint are eliminated by morphological closing operations of 5×5 structuring elements to smooth the imprint boundary.

[0049] Based on the above embodiments, this embodiment calculates the actual area of ​​the imprint based on the optimized target area image, including: detecting the imprint contour in the optimized target area image based on the Suzuki85 boundary tracking algorithm, correcting local abnormal protrusions or depressions of the imprint contour using a contour optimization algorithm (such as the rolling circle algorithm) to obtain a regular imprint contour; counting the number of pixels in the regular imprint contour, determining the conversion relationship between pixels and actual length in combination with the resolution parameters of the imprint image acquisition device, and calculating the actual area of ​​the imprint.

[0050] For example, the number of pixels n within the optimized outline is counted, and the conversion factor is calibrated based on a 300 dpi resolution. That is, 1 pixel corresponds to an actual area of ​​0.007213 mm², which is determined by the formula. Calculate the actual area of ​​the imprint (Unit: cm²)

[0051] The maximum flight speed of the projectile was measured using a velocimeter (maximum permissible error < 0.0001 sec). (Unit: m / s) The mass of the projectile is measured using an electronic weighing device (accuracy 0.001 g). (Unit: kg); The speedometer and weighing equipment are connected to the data processing system via USB data interface to achieve automatic data reading and avoid manual input errors. The kinetic energy calculation formula is: The formula for calculating the kinetic energy per unit contact area is: .

[0052] Pass / Fail Judgment: If If it is deemed qualified; Then it must satisfy Furthermore, a warning explanation must be attached; otherwise, it will be deemed unqualified. Figure 6 is a visualization diagram of the contour extraction and area calculation of circular imprint samples.

[0053] Based on the above embodiments, this embodiment, after inputting the imprint image generated by the impact of the projectile launched by the catapult and the projectile's associated information into the improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image output by the improved target detection model, further includes: if the proportion of missing imprint edges within the imprint prediction box exceeds a preset missing threshold, then the imprint is deemed unqualified, and the imprint prediction box and the corresponding imprint prediction category are removed; if the imprint prediction category is a circular imprint, then the ellipticity of the circular imprint is calculated; if the imprint prediction category is a concentric ring imprint, then the interlayer spacing of the concentric ring imprint is calculated; if the ellipticity or the interlayer spacing exceeds a preset deviation threshold for the corresponding imprint prediction category, then the imprint is deemed unqualified, and the imprint prediction box and the corresponding imprint prediction category are removed.

[0054] An improved target detection model is used for detection, outputting predicted imprint bounding boxes and corresponding predicted imprint categories to locate target regions. An automatic identification and removal of unqualified imprints is achieved through a multi-dimensional judgment mechanism based on imprint integrity and morphological regularity.

[0055] For example, regarding the integrity rules of the imprint, if the proportion of missing imprint edges is greater than 10%, it is judged as unqualified; regarding the regularity of shape, if the ellipticity of a circular imprint is greater than 0.2, or the deviation of the interlayer spacing of concentric ring imprints is greater than 0.5 mm, it is judged as unqualified; finally, only the valid target area is retained for subsequent area calculation.

[0056] Based on the YOLOv12 model, and considering the detection characteristics of projectile toy imprints that are "small in size, multi in shape (circular / concentric ring), and high in contrast", as shown in Figure 3, improvements were made from two dimensions: attention module fusion and network lightweight adaptation, ultimately forming the ASC-YOLOv12 model.

[0057] Base model selection: The performance parameters of five versions of the YOLOv12 base model are compared, as shown in Table 1.

[0058] Table 1 Performance Comparison of 5 Versions of YOLOv12 Model

[0059] This embodiment selects YOLOv12 versions N and S as the improvement baseline. YOLOv12-N version has the advantage of being lightweight, specifically with 2.52 × 10^6 parameters. 6 With a computational cost of 6.0 GFLOPs, it is suitable for deployment on edge devices; the YOLOv12-S version offers superior detection accuracy, specifically with 9.10 × 10^6 parameters. 6With a computational cost of 19.6 GFLOPs, it is suitable for scenarios with higher accuracy requirements. Both retain the original multi-scale feature fusion mechanism of YOLOv12 (to capture imprints at different scales through feature pyramids) and anchor box optimization strategy (to adapt to the circular / ring contours of imprints), and only enhance the feature extraction capability.

[0060] Module Comparison Analysis: The compatibility of the CBAM and SimAM attention modules was tested separately. The CBAM module enhances target features through a dual-dimensional weighted approach of "channel attention + spatial attention," but this increases network computational complexity, raising the parameter count of the CBAM-YOLOv12-N version to 3.42 × 10⁻⁶. 6 The computational cost increases to 10.5 GFLOPs; the SimAM module achieves adaptive feature weighting through simple nonlinear operations, without introducing additional parameters, and can improve performance while maintaining the original YOLOv12 network size. That is, the SinAM-YOLOv12-N version still has 2.55 × 10^6 parameters. 6 The computational cost is 6.5 GFLOPs.

[0061] Combination Module Construction: To balance accuracy and lightweight design, an A2C2f-SimAM-CBAM combined attention structure, abbreviated as ASC structure, is designed: The first layer embeds the A2C2f module, which expands the network's receptive field and improves the feature capture efficiency for imprints of different sizes (such as small circles with a diameter of 5mm and concentric rings with a diameter of 20mm) through adaptive channel and spatial attention mechanisms; the second layer embeds the SimAM module, which adaptively weights the feature map output by the A2C2f module in a lightweight manner, strengthening key features such as imprint edges and textures, while avoiding a surge in computational cost; the third layer embeds the CBAM module, which supplements channel attention and spatial attention, highlighting feature channels with significant differences between the imprint and the background, and focusing on the core region of the imprint, solving the problems of interlayer interference in concentric ring imprints and the lack of obvious features in small-sized imprints.

[0062] As shown in Figure 5, this combined structure is embedded into the backbone feature extraction layer of the YOLOv12 model to form the ASC-YOLOv12 model. The improved model is trained and its performance is validated using the imprint dataset.

[0063] Ablation experiments and performance verification: The effectiveness of each improved module was verified through ablation experiments. The performance of five schemes, namely the basic YOLOv12, YOLOv12+CBAM, YOLOv12+SimAM, YOLOv12+A2C2f-SimAM, and YOLOv12+A2C2f-SimAM-CBAM (ASC-YOLOv12), was tested. The key indicators are shown in Table 2.

[0064] Experimental results show that the ASC-YOLOv12 model (N / S versions) maintains high precision (≥98%) and recall (≥99.7%), while significantly outperforming the basic YOLOv12 and single attention module improvement schemes in terms of mAP@0.5:0.95. Specifically, the S version achieves 87.0% mAP@0.5:0.95 and can accurately identify circular and concentric ring imprints; the N version has only 3.20 × 10^6 parameters. 6 It balances lightweight design with testing accuracy.

[0065] This embodiment solves the problems of insufficient accuracy and excessive computation of the basic YOLOv12 model in recognizing multi-form imprints on projectile toys by adopting the improved strategy of "multi-attention module combination + lightweight adaptation". The constructed ASC-YOLOv12 model can be efficiently adapted to imprint detection scenarios.

[0066] Table 2. Performance Comparison of Each Improved Module and Combination Verified by Ablation Experiments

[0067] This embodiment focuses on verifying the accuracy of measuring the area of ​​projectile toy imprints. It verifies the accuracy and error control capability of the measurement method of this application by comparing the calibration of a standard circular sample with the actual measurement of a circular imprint sample.

[0068] To eliminate systematic errors in the "image acquisition, contour extraction, and area calculation" stages of the measurement process, a standard circular sample with a known theoretical area was used for calibration testing. The specific process is as follows: Preparation of standard circular sample: A standard circular pattern with a diameter of 10 mm was printed using a high-precision printer, and the theoretical area was calculated according to the formula for the area of ​​a circle. Five sets of standard circular samples were prepared, numbered from sample 0 to sample 4, ensuring that each set of samples was free of edge defects and printing deformation.

[0069] Image Acquisition and Processing: An AURORA AD369s flatbed scanner with a resolution of 300 dpi was used to acquire digital images of standard circular samples. The images were processed according to the image processing workflow in Example 1. The contours were extracted and optimized using the Suzuki85 boundary tracking algorithm and the rolling circle algorithm. The number of pixels within the contours was counted, and a conversion factor calibrated for 300 dpi resolution was used. Calculate the actual measured area.

[0070] Accuracy and Error Analysis: The measured areas of 5 sets of standard circular samples were recorded and compared with the theoretical areas to calculate the error. The results are shown in Table 3.

[0071] Table 3 Comparison of theoretical and measured values ​​of the area of ​​standard circular samples

[0072] As shown in the table above, the average measurement value of the five standard circular samples is 80.3 mm², with a relative error of 2.2% compared to the theoretical area of ​​78.54 mm², and a standard deviation of only 0.44 mm². The error mainly comes from minor dimensional deviations during the printing process and slight uneven lighting during scanning, which is within the acceptable range of systematic error. This verifies the stability and accuracy of this application in the "image processing and area calculation" process.

[0073] To further verify the applicability of the method in actual testing scenarios, a total of 10 actual circular imprint samples from three batches (numbered B1-B5, A1-A3, and H1-H2) were selected. The results of manual measurement and automatic measurement in this application were compared to analyze the actual measurement accuracy of the method. The specific process is as follows: Preparation of actual imprint samples: Using a circular projectile toy, 10 circular impact imprints were generated according to GB 6675.2-2014 standard (the projectile was coated with red pigment and formed by impacting white paper). Among them, sample A1 was judged as unqualified because the edge of the imprint was deformed due to ink overflow. Only the area of ​​the remaining 9 qualified samples was compared.

[0074] Manual measurement method: The traditional grid paper counting method (grid size 1mm×1mm) is adopted. Three professional inspectors independently count the number of complete squares covered by each imprint. For incomplete squares, the "rounding" principle is used to estimate the area. The average of the three people's measurement results is taken as the area measured manually.

[0075] Automatic measurement method: The target area of ​​the imprint is located using the ASC-YOLOv12 model. After image processing and contour extraction, the area is automatically measured, and the absolute error is recorded. With relative error .

[0076] Comparison of measured results: The measurement results of 9 qualified circular imprint samples are shown in Table 4. The automatically measured area of ​​the 9 qualified circular imprint samples is highly consistent with the manually measured area, with an average absolute error of 1.0 mm² and an average relative error of only 0.95%, which is controlled within 1%, significantly better than the error range of traditional manual measurement. The largest relative error occurred in sample B3 (1.65%), mainly because there was slight pigment smudging at the edge of this sample. The judgment of the smudging area during manual measurement has a subjective bias, while the automatic measurement accurately eliminated the invalid smudging area through the dual feature screening of "grayscale gradient + edge curvature", which is closer to the actual effective imprint area. The unqualified sample A1 was automatically eliminated by the ASC-YOLOv12 model due to edge deformation and was not included in the calculation, avoiding the interference of invalid data on the detection results.

[0077] Table 4 Comparison of Measured Areas of Circular Imprint Samples

[0078] This embodiment verifies the high accuracy of the measurement method of this application by comparing the calibration of standard circular samples with the actual measurement of actual imprint samples. The relative error can be controlled within 1%, and it has the ability to automatically identify unqualified imprints. This provides key data support for the reliability of subsequent calculation of kinetic energy and kinetic energy per unit area, and solves the problems of poor accuracy and large subjective interference in traditional manual measurement.

[0079] The catapult kinetic energy measuring device provided by the present invention is described below. The catapult kinetic energy measuring device described below and the catapult kinetic energy measuring device method described above can be referred to in correspondence with each other.

[0080] As shown in Figure 7, the device includes an imprint prediction module 701, an area calculation module 702, and a kinetic energy calculation module 703. The imprint prediction module 701 inputs the imprint image generated by the impact of a projectile launched by the catapult and the projectile's warhead association information into an improved target detection model to obtain the imprint prediction bounding box and imprint prediction category in the imprint image output by the improved target detection model. The area calculation module 702 optimizes the target region image within the imprint prediction bounding box and calculates the actual imprint area based on the optimized target region image. The kinetic energy calculation module 703 calculates the kinetic energy of the projectile based on its mass and post-launch flight speed, and calculates the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual imprint area. The improved target detection model is trained by using imprint image samples as samples and the imprint category labeling results and imprint bounding box labeling results in the imprint image samples as labels.

[0081] This embodiment addresses the kinetic energy detection requirements of catapults. Taking catapults with different projectiles as the detection objects, it employs an improved target detection model to accurately identify different types of projectiles based on the imprint images generated by the projectiles and the projectile association information. Combined with image optimization processing technology, it ensures the accuracy of imprint area measurement, thereby achieving rapid and accurate measurement of catapult kinetic energy. The fully automated design significantly improves detection efficiency.

[0082] Figure 8 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 8, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logic instructions in the memory 830 to execute a catapult kinetic energy measurement method. This method includes: inputting an image of the imprint generated by the impact of a projectile launched by the catapult and the projectile's warhead association information into an improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image output by the improved target detection model; optimizing the target region image within the imprint prediction box and calculating the actual imprint area based on the optimized target region image; calculating the kinetic energy of the projectile based on its mass and post-launch flight speed, and calculating the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual imprint area; the improved target detection model is trained by using imprint image samples and corresponding projectile association information as samples, and using the imprint category labeling results and imprint box labeling results in the imprint image samples as labels.

[0083] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the catapult kinetic energy measurement method provided by the above methods. The method includes: inputting an imprint image generated by the impact of a projectile launched by the catapult and the projectile's warhead association information into an improved target detection model to obtain an imprint prediction box and an imprint prediction category in the imprint image output by the improved target detection model; optimizing the target region image within the imprint prediction box and calculating the actual area of ​​the imprint based on the optimized target region image; calculating the kinetic energy of the projectile based on its mass and flight speed after launch, and calculating the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual area of ​​the imprint; the improved target detection model is obtained by training the imprint image samples and the corresponding warhead association information as samples, and using the imprint category labeling results and imprint box labeling results in the imprint image samples as labels.

[0085] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program performs the catapult kinetic energy measurement method provided by the above methods. The method includes: inputting an imprint image generated by the impact of a projectile launched by the catapult and the projectile's warhead association information into an improved target detection model to obtain an imprint prediction box and an imprint prediction category in the imprint image output by the improved target detection model; optimizing the target region image within the imprint prediction box and calculating the actual imprint area based on the optimized target region image; calculating the kinetic energy of the projectile based on its mass and flight speed after launch, and calculating the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual imprint area; the improved target detection model is obtained by training the imprint image samples and the corresponding warhead association information as samples, and using the imprint category labeling results and imprint box labeling results in the imprint image samples as labels.

[0086] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0087] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for measuring the kinetic energy of a catapult, characterized in that, include: The imprint image generated by the impact of the projectile launched by the catapult and the associated information of the projectile's warhead are input into the improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image output by the improved target detection model. After optimizing the target region image within the imprint prediction box, the actual imprint area is calculated based on the optimized target region image; the kinetic energy of the projectile is calculated based on its mass and flight speed after launch, and the kinetic energy per unit area of ​​the projectile is calculated based on its kinetic energy and the actual imprint area. The improved target detection model is trained by using imprint image samples and corresponding warhead association information as samples, and imprint category annotation results and imprint box annotation results in the imprint image samples as labels.

2. The catapult kinetic energy measurement method according to claim 1, characterized in that, The improved object detection model is based on the YOLOv12 model. A multi-module collaborative attention mechanism combination structure is embedded in the backbone network of the YOLOv12 model. The attention mechanism combination structure includes an adaptive feature enhancement module, a lightweight feature weighting module, and a multi-dimensional feature enhancement module.

3. The catapult kinetic energy measurement method according to claim 2, characterized in that, The backbone network includes a first combined structure of a first C3K2 module, an adaptive feature enhancement module, and a lightweight feature weighting module, and a second combined structure of an adaptive feature enhancement module and a multi-dimensional feature enhancement module; the first C3K2 module, the first combined structure, and the second combined structure are connected sequentially; the neck network of the YOLOv12 model includes a third combined structure, a fourth combined structure, and a fifth combined structure of a second C3K2 module, an adaptive feature enhancement module, and a lightweight feature weighting module; the output of the second combined structure is upsampled and concatenated with the output of the first combined structure, and the concatenation result is input into the third combined structure; the output of the third combined structure is upsampled and concatenated with the output of the first C3K2 module, and the concatenation result is input into the fourth combined structure; the output of the fourth combined structure is input into the first detection module in the head network of the YOLOv12 model; the output of the fourth combined structure is concatenated with the output of the third combined structure and input into the fifth combined structure; the output of the fifth combined structure is input into the second detection module in the head network; The output of the fifth combined structure is concatenated with the output of the second combined structure and then input into the second C3K2 module. The output of the second C3K2 module is input into the third detection module in the head network.

4. The catapult kinetic energy measurement method according to claim 1, characterized in that, Before inputting the imprint image generated by the impact of the projectile launched by the catapult and the projectile's warhead association information into the improved target detection model, the method further includes: inputting the imprint image sample and the corresponding warhead association information into the improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image sample output by the improved target detection model; calculating the classification loss between the imprint prediction category in the imprint image sample and the corresponding imprint category labeling result, and the bounding box regression loss between the imprint prediction box in the imprint image sample and the corresponding imprint box labeling result; obtaining the target loss function based on the component loss and the bounding box regression loss; updating the parameters of the improved target detection model through backpropagation based on the target loss function until the change value of the target loss function satisfies the preset convergence condition or the number of training iterations reaches the preset threshold.

5. The catapult kinetic energy measurement method according to claim 1, characterized in that, The optimization processing of the target region image within the imprint prediction box includes: converting the target region image into a grayscale image and then using a smoothing algorithm to suppress noise, thereby reducing interference noise introduced during image acquisition; performing binarization on the target region image using an adaptive thresholding strategy to distinguish the imprint region from the background region and highlight the imprint contour features; removing isolated noise points in the target region image through morphological opening operations and eliminating small holes inside the imprint in the target region image by combining morphological closing operations, thereby optimizing the imprint boundary morphology.

6. The catapult kinetic energy measurement method according to claim 1, characterized in that, The actual area of ​​the imprint is calculated based on the optimized target area image, including: detecting the imprint contour in the optimized target area image based on the Suzuki85 boundary tracking algorithm; using a contour optimization algorithm to correct local abnormal protrusions or depressions in the imprint contour to obtain a regular imprint contour; counting the number of pixels in the regular imprint contour; determining the conversion relationship between pixels and actual length based on the resolution parameters of the imprint image acquisition device; and calculating the actual area of ​​the imprint.

7. The catapult kinetic energy measurement method according to claim 1, characterized in that, After inputting the impact image of the projectile launched by the catapult and the projectile's associated information into the improved target detection model to obtain the predicted bounding box and predicted category of the projectile in the projectile image output by the improved target detection model, the method further includes: if the proportion of missing edges of the projectile within the predicted bounding box exceeds a preset missing threshold, the projectile is deemed unqualified, and the predicted bounding box and the corresponding predicted category are removed; if the predicted category is a circular projectile, the ellipticity of the circular projectile is calculated; if the predicted category is a concentric ring projectile, the interlayer spacing of the concentric ring projectile is calculated; if the ellipticity or the interlayer spacing exceeds a preset deviation threshold for the corresponding predicted category, the projectile is deemed unqualified, and the predicted bounding box and the corresponding predicted category are removed.

8. A catapult kinetic energy measuring device, characterized in that, include: The imprint prediction module is used to input the imprint image generated by the impact of the projectile launched by the catapult and the warhead association information of the projectile into the improved target detection model to obtain the imprint prediction box and imprint prediction category in the imprint image output by the improved target detection model. The area calculation module is used to optimize the target area image within the imprint prediction box and then calculate the actual area of ​​the imprint based on the optimized target area image. The kinetic energy calculation module is used to calculate the kinetic energy of the projectile based on its mass and flight speed after launch, and to calculate the kinetic energy per unit area of ​​the projectile based on its kinetic energy and the actual area of ​​the imprint. The improved target detection model is trained by using imprint image samples as samples and imprint category annotation results and imprint bounding box annotation results in the imprint image samples as labels.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the catapult kinetic energy measurement method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the catapult kinetic energy measurement method as described in any one of claims 1 to 7.

Images