Automatic sorting method and system for surface flaws of fire extinguisher tank based on machine vision

Abstract

The invention provides a fire extinguisher tank surface flaw automatic sorting method and system based on machine vision. According to the method, an infrared thermal imager and a controllable annular array halogen lamp group are adopted to apply uniform heat flow to the surface of a fire extinguisher tank body, and a surface thermal imaging temperature change sequence is collected to generate thermal imaging video stream data; the thermal response difference between the flaws and the normal area is distinguished by modulating the frequency and power of the thermal excitation source; performing time sequence thermal image analysis on the thermal image video stream data to obtain a temperature attenuation curve and a phase diagram, and converting the temperature attenuation curve and the phase diagram into frequency domain features through Fourier transform; analyzing a thermal response difference by using the frequency domain characteristics, and identifying a thermal signal difference region; the difference areas are segmented, defect types are classified through morphological processing, and sorting signals are output to achieve automatic sorting. Precise recognition and sorting of the surface flaws of the fire extinguisher tank are achieved, the omission ratio is reduced, and the sorting efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of automatic defect sorting technology, and in particular to a method and system for automatic sorting of defects on the surface of fire extinguisher canisters based on machine vision. Background Technology

[0002] As a pressure-bearing safety component, the surface defects of fire extinguisher canisters directly affect the product's sealing performance and explosion-proof reliability. In automated production lines, traditional manual visual inspection struggles to meet the dual challenges of high-speed production lines (inspecting dozens of canisters per minute) and identifying microscopic defects (such as millimeter-level microcracks). Furthermore, the cylindrical curved surface of the canisters causes strong optical reflection interference, and the morphology of defects is highly random (e.g., differences in the appearance of welding spatter and mechanical impacts), necessitating multi-angle, high-precision visual coverage and adaptive feature extraction. In addition, the surface reflectivity and coating color differences of canisters made of different materials (stainless steel / carbon steel) further increase the complexity of inspection. Therefore, a fully automated sorting system compatible with multiple canister types and capable of identifying defect types and levels is urgently needed to prevent defective canisters from flowing into subsequent processes and causing safety hazards.

[0003] Current solutions are based on machine vision sorting systems using multi-station collaborative imaging. These systems capture images of the can surface from multiple angles using a ring of high-resolution industrial cameras, combined with polarization filtering to suppress metallic reflection interference. The image processing module employs a convolutional neural network to extract and classify defect features hierarchically (e.g., distinguishing between rust and scratches), and works in conjunction with a motion control unit and robotic arm to sort defective cans to different recycling channels. However, its core weakness lies in its still relatively high rate of missed detections in highly reflective curved areas, and the model's generalization ability is limited by the training data, making it difficult to adapt to entirely new can materials or rare defect patterns. Summary of the Invention

[0004] This application provides a machine vision-based automatic sorting method and system for surface defects of fire extinguisher canisters, which solves the problems of low efficiency, insufficient thermal excitation response differentiation, and high rate of missed detection of minor defects in the prior art.

[0005] In a first aspect, this application provides a machine vision-based method for automatically sorting surface defects on fire extinguisher canisters, including: An infrared thermal imager and a controllable ring array of halogen lamps are used as thermal excitation sources to apply a uniform heat flow to the surface of the fire extinguisher tank. At the same time, the change sequence of thermal imaging temperature data on the surface of the fire extinguisher tank is collected to generate thermal imaging video stream data. The difference in thermal response between defective areas and normal areas on the surface of the fire extinguisher canister is distinguished by modulating the frequency and power of the thermal excitation source. Temporal thermal imaging analysis is performed on the thermal video stream data to obtain the temperature decay curve and phase map in the thermal video stream data as temporal temperature data, and the temporal temperature data is converted into frequency domain features through Fourier transform. The thermal response differences are analyzed using the frequency domain features to identify the areas of thermal signal difference between defective and normal areas; The thermal signal difference region is segmented, and the segmented region is classified into defect types through morphological processing to output a sorting signal, and automatic sorting is performed according to the sorting signal.

[0006] Optionally, an infrared thermal imager and a controllable ring array halogen lamp group are used as thermal excitation sources to apply a uniform heat flow to the surface of the fire extinguisher tank, while simultaneously collecting the change sequence of thermal imaging temperature data of the fire extinguisher tank surface to generate thermal video stream data, including: A controllable ring array of halogen lamps is installed in a ring above the fire extinguisher tank conveyor line, and the illumination angle of the controllable ring array of halogen lamps covers the surface of the fire extinguisher tank and the heat flow is evenly distributed. An infrared thermal imager is installed at the controllable ring array halogen lamp group, and the focal length of the infrared thermal imager is adjusted so that the field of view of the infrared thermal imager completely covers the surface area of ​​the fire extinguisher canister being tested. When the fire extinguisher tank enters the inspection station, the controllable ring array halogen lamp group is activated to emit heat flow and trigger the infrared thermal imager to record the temperature data of the fire extinguisher tank surface at a fixed acquisition frequency to form a time-sequenced thermal imaging temperature data change sequence. The change sequence of the thermal imaging temperature data is integrated in chronological order to generate thermal video stream data.

[0007] Optionally, the thermal signal difference region is segmented, and the segmented region is classified into defect types through morphological processing to output a sorting signal. Automatic sorting is then performed based on the sorting signal, including: Acquire the thermal signal difference value of the thermal signal difference region, and mark the region where the thermal signal difference value exceeds the preset segmentation threshold as the foreground region, and the region where the thermal signal difference value does not exceed the preset segmentation threshold as the background region; Connect adjacent foreground regions and fill the internal voids of the foreground regions to form connected regions; Extract the morphological feature parameters of the connected regions, match the morphological feature parameters with preset defect type features, and determine the defect type identifier corresponding to each connected region based on the matching result; A sorting instruction signal is generated based on the defect type identifier, and the sorting instruction signal is sent to the automatic sorting device to perform the sorting operation.

[0008] Optionally, morphological feature parameters of the connected regions are extracted, and the morphological feature parameters are matched with preset defect type features. Based on the matching results, a defect type identifier corresponding to each connected region is determined, including: Measure the area of ​​each connected region and calculate the ratio of the length to the width of the bounding rectangle of the connected region as the aspect ratio feature value; The ratio of the number of boundary pixels of a connected region to the square of the perimeter of the connected region is used as the boundary complexity feature value. The region area value, aspect ratio feature value, and boundary complexity feature value are combined into morphological feature parameters; The morphological feature parameters are compared with a preset defect type feature threshold to generate a successful match signal; The defect type identifier of the connected region is determined based on the defect type feature threshold corresponding to the successful matching signal.

[0009] Optionally, the thermal response differences are analyzed using the frequency domain features to identify regions with different thermal signals between defective and normal regions, including: Extract the frequency domain amplitude feature value and frequency domain phase feature value of each spatial location point from the frequency domain features; The thermal response difference is obtained at the corresponding spatial location point in the thermal response difference, and the thermal response difference describes the degree of difference in thermal conductivity between the defective area and the normal area. The frequency domain amplitude feature value is combined with the thermal response difference to generate a comprehensive signal strength value, and the comprehensive signal strength value is compared with a preset normal area signal strength threshold. When the comprehensive signal strength value exceeds the normal area signal strength threshold, an anomaly marker signal for the spatial location point is generated. The abnormal marker signals of all spatial locations are integrated and arranged in spatial coordinate order to form a thermal signal difference region that characterizes the difference between the defective area and the normal area.

[0010] Optionally, time-series thermal imaging analysis is performed on the thermal video stream data to obtain the temperature decay curve and phase map in the thermal video stream data as time-series temperature data, and the time-series temperature data is converted into frequency domain features through Fourier transform, including: Extract the temperature time series of each spatial location point from the thermal imaging video stream data, the temperature time series including the temperature measurement value of each time point; The temperature time series is subjected to exponential decay curve fitting to generate a temperature decay curve describing the temperature change over time. Calculate the phase difference between adjacent time points in the temperature time series, and connect the phase differences in chronological order to form a phase diagram; The temperature decay curve and phase diagram are integrated into time-series temperature data, and the variation characteristics of the time-series temperature data are obtained. The change features are converted into frequency domain amplitude feature values ​​and frequency domain phase feature values, and the frequency domain amplitude feature values ​​and frequency domain phase feature values ​​are combined to form frequency domain features.

[0011] Optionally, the measurable difference in thermal response between defective areas and normal areas on the surface of the fire extinguisher canister due to different thermal conductivity can be distinguished by modulating the frequency and power of the thermal excitation source, including: Set the basic frequency and basic power parameters of the thermal excitation source, and adjust the frequency of the thermal excitation source from the basic frequency to the first frequency pulse mode, while increasing the power from the basic power to the first power level; At the first power level, thermal imaging temperature data of the surface of the fire extinguisher tank is collected, and the instantaneous change characteristics of the thermal imaging temperature data are recorded. The frequency of the thermal excitation source is restored from the first frequency pulse mode to the fundamental frequency, while the power is reduced from the first power level to the fundamental power. Thermal imaging temperature data of the surface of the fire extinguisher tank was collected under the basic frequency and basic power modes, and the steady-state characteristics of the thermal imaging temperature data were recorded. The instantaneous change characteristics are compared with the steady-state characteristics to generate thermal response differences.

[0012] Secondly, this application provides an automatic sorting system for surface defects of fire extinguisher canisters based on machine vision, comprising: The acquisition module is used to apply a uniform heat flow to the surface of the fire extinguisher tank using an infrared thermal imager and a controllable ring array halogen lamp group as thermal excitation sources, and at the same time acquire the change sequence of thermal imaging temperature data of the surface of the fire extinguisher tank to generate thermal imaging video stream data. The differentiation module is used to distinguish between the defective areas and normal areas on the surface of the fire extinguisher canister by modulating the frequency and power of the thermal excitation source to create measurable differences in thermal response due to differences in thermal conductivity. The conversion module is used to perform time-series thermal imaging analysis on the thermal video stream data to obtain the temperature decay curve and phase map in the thermal video stream data as time-series temperature data, and to convert the time-series temperature data into frequency domain features through Fourier transform. The identification module is used to analyze the thermal response differences using the frequency domain features to identify the thermal signal difference areas between defective areas and normal areas. The output module is used to segment the thermal signal difference area, classify the defect type of the segmented area through morphological processing, output a sorting signal, and perform automatic sorting according to the sorting signal.

[0013] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement an automatic sorting method for surface defects of fire extinguisher canisters based on machine vision as described in the first aspect above.

[0014] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a machine vision-based automatic sorting method for surface defects of fire extinguisher canisters as described in the first aspect.

[0015] This application's technical solution achieves high-precision detection of surface defects on fire extinguisher canisters through active thermal excitation and intelligent frequency domain analysis technology. Specifically, controllable heat source modulation and synchronous infrared thermography significantly improve the ability to identify differences in thermal response between defects and the substrate; frequency domain conversion of time-series temperature data effectively enhances the detection sensitivity of deep defects; and morphological classification algorithms enable automated identification and sorting of defect types. This method overcomes the limitations of traditional single thermal imaging detection, forming a fully automated detection system from thermal excitation to intelligent sorting, significantly improving the efficiency and accuracy of fire extinguisher canister surface quality control.

[0016] Furthermore, a highly efficient and automated method for detecting surface defects in fire extinguisher canisters is achieved through ring-shaped thermal excitation and synchronous infrared acquisition technology. Specifically, the uniform heat flow application based on a controllable halogen lamp significantly improves the consistency of surface thermal response; the full-coverage field-of-view configuration of the infrared thermal imager ensures the integrity and continuity of temperature data acquisition; and the integration mechanism of time-series temperature sequences provides a high-fidelity data foundation for subsequent frequency domain analysis. This method overcomes the limitations of traditional manual inspection, achieving precise coordination between thermal excitation and image acquisition, providing a reliable thermal imaging data source for automated sorting systems, and significantly improving the efficiency and accuracy of fire extinguisher canister surface quality inspection.

[0017] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application 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 application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of an automatic sorting method for surface defects of fire extinguisher canisters based on machine vision, provided in this application, is shown. Figure 2 A schematic diagram of the structure of an automatic sorting system for surface defects of fire extinguisher canisters based on machine vision provided in this application is shown. Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0021] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0022] Researchers have discovered a fundamental bottleneck in fire extinguisher canister surface defect detection technology: while multi-station collaborative imaging-based machine vision solutions can achieve multi-angle image acquisition, their insufficient suppression of reflective interference and lack of material generalization ability lead to dual detection failures. Specifically, optical reflection noise in highly reflective curved areas drowns out micron-level defect features, and limited training data causes a surge in the model's false negative rate for canisters made of novel materials or with rare defect patterns. This contradiction stems from the inability to perceive differences in surface thermophysical properties and the inherent limitations of optical imaging, necessitating the construction of a cross-modal detection architecture that links thermal excitation and thermal response.

[0023] To address the aforementioned challenges, this invention proposes an automatic sorting method for surface defects on fire extinguisher canisters based on machine vision. Its innovation lies in overcoming the limitations of optical vision detection through heat flux excitation modulation and temporal thermal imaging analysis. Specifically: a controllable ring-shaped halogen lamp group applies uniform heat flux excitation to the canister surface, while simultaneously acquiring a thermal imaging temperature change sequence using an infrared thermal imager and generating thermal video stream data; by modulating the frequency and power of the heat excitation source, the difference in thermal response between defective and normal areas due to differences in thermal conductivity is amplified; temporal analysis of the thermal video stream data is performed to obtain temperature decay curves and phase maps, which are then interpreted into frequency domain features via Fourier transform; based on the frequency domain features, regions with different thermal signals are identified, and morphological processing is used to classify defects and trigger sorting instructions. This method overturns the limitations of traditional optics: thermal excitation frequency domain modulation achieves millisecond-level thermal response capture of microcracks on the surface of tanks of different materials for the first time, completely solving the problem of feature submersion caused by reflective interference; time-series thermal imaging analysis maintains detection stability between carbon steel and stainless steel tanks by fusing temperature attenuation and phase features; morphological classification engine transforms thermal signal differences into sorting decisions, forming a closed-loop detection chain of "thermal excitation-response acquisition-frequency domain interpretation-sorting execution", providing pressure vessels with end-to-end quality assurance from thermophysical properties to automatic sorting.

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Figure 1 This application provides a flowchart of a machine vision-based automatic sorting method for surface defects on fire extinguisher canisters, as shown in the embodiments below. Figure 1 As shown, the method includes: 101. An infrared thermal imager and a controllable ring array halogen lamp group are used as thermal excitation sources to apply a uniform heat flow to the surface of the fire extinguisher tank, and at the same time, the change sequence of thermal imaging temperature data of the surface of the fire extinguisher tank is collected to generate thermal imaging video stream data.

[0026] Optionally, step 101 may specifically include the following steps: 1011. A controllable annular array halogen lamp group is installed in a ring above the fire extinguisher tank conveyor line, and the illumination angle of the controllable annular array halogen lamp group covers the surface of the fire extinguisher tank and the heat flow is evenly distributed.

[0027] 1012. Install an infrared thermal imager at the controllable ring array halogen lamp group, and adjust the focal length of the infrared thermal imager so that the field of view of the infrared thermal imager completely covers the surface area of ​​the fire extinguisher canister being tested.

[0028] 1013. When the fire extinguisher tank enters the inspection station, the controllable ring array halogen lamp group is activated to emit heat flow and trigger the infrared thermal imager to record the temperature data of the surface of the fire extinguisher tank at a fixed acquisition frequency to form a time-sequenced sequence of thermal imaging temperature data changes.

[0029] 1014. The change sequence of the thermal imaging temperature data is integrated in chronological order to generate thermal imaging video stream data.

[0030] In the above scheme, an infrared thermal imager refers to an instrument for infrared thermal imaging. A controllable ring array halogen lamp group refers to a controllable ring array of halogen lamps. A thermal excitation source refers to a source that provides thermal excitation. Heat flow refers to the flow of heat. Thermal imaging temperature data refers to the temperature data from thermal imaging. Thermal imaging video stream data refers to the video stream data from thermal imaging. A fire extinguisher canister conveyor line refers to a production line that conveys fire extinguisher canisters. Illumination angle refers to the angle at which the light illuminates the canister. Field of view refers to the range of the field of view. Inspection station refers to the work position for inspection. Acquisition frequency refers to the frequency of data acquisition. Time sequence refers to the order in which data is arranged chronologically.

[0031] In this embodiment, firstly, in step 1011, a controllable annular array of halogen lamps is installed in a ring above the fire extinguisher tank conveyor line, ensuring that the illumination angle of the controllable annular array of halogen lamps covers the surface of the fire extinguisher tank and that the heat flow is evenly distributed. This process uses mechanical supports to fix multiple halogen lamps in a ring array directly above the conveyor line. By adjusting the pitch and azimuth angles of each lamp, their beams converge and completely cover the tank surface. Simultaneously, based on the principle of light field uniformity, the lamp spacing and illumination angle are determined through simulation calculations, ensuring that the superimposed heat flow is distributed evenly on the tank surface, forming a uniform energy coverage and providing stable and dead-angle-free heat input conditions for subsequent thermal excitation.

[0032] Subsequently, in step 1012, an infrared thermal imager is installed at the controllable ring array halogen lamp assembly, and the focal length of the infrared thermal imager is adjusted to ensure that the field of view of the infrared thermal imager completely covers the surface area of ​​the fire extinguisher canister being tested. The system fixes the infrared thermal imager to the center of the lamp assembly or to a side bracket, and manually focuses or activates the autofocus function by rotating its focal ring to keep the optical center of the lens perpendicularly aligned with the surface of the canister; the field of view is calculated according to the canister size and detection distance, and the focal length is adjusted until the edge of the canister is clear and distortion-free in the thermal imager display screen, ensuring that the thermal radiation signal of the entire surface area can be completely acquired.

[0033] Next, in step 1013, when the fire extinguisher tank enters the detection station, the controllable ring array halogen lamp group is activated to emit heat flux and triggers the infrared thermal imager to record the temperature data of the fire extinguisher tank surface at a fixed acquisition frequency, forming a time-sequential sequence of thermal imaging temperature data changes. After the photoelectric sensor detects that the tank has arrived at the station, the control system synchronously sends a trigger signal to the halogen lamp group and the thermal imager; after receiving the command, the halogen lamp group emits a constant heat flux, while the thermal imager continuously captures infrared radiation on the tank surface at a preset frame rate; its detector converts the radiation energy into an electrical signal, which, after amplification and analog-to-digital processing, generates continuous temperature matrix data and stacks it in time-stamp order to form a three-dimensional temperature change sequence.

[0034] Finally, in step 1014, the change sequence of thermal imaging temperature data is integrated in chronological order to generate thermal video stream data. The system uses the temperature matrix data at each time point as video frames, and uses an encoding algorithm to map the frame rate, resolution, and temperature value into grayscale or pseudo-color pixels; then, it encapsulates multiple frames of data into a standard video container format in chronological order, where the pixel value of each frame corresponds to the actual temperature distribution, ultimately generating thermal video stream data that can be played in real time or analyzed later.

[0035] In practical applications, on the automatic sorting production line for surface defects of fire extinguisher canisters, the system first installs a controllable ring array of halogen lamps in a circular pattern above the fire extinguisher canister conveyor line. The lamps are adjusted to ensure that the heat flow covers the entire surface of the canister and that the heat flow is evenly distributed (avoiding temperature distortion caused by local overheating or underheating). Then, an infrared thermal imager (which supports long-wave infrared band acquisition) is installed at the midpoint of the controllable ring array of halogen lamps. The focal length of the infrared thermal imager is precisely adjusted so that its field of view completely covers the surface area of ​​the fire extinguisher canister being tested (including the canister busbar, end cap transition area, and other areas prone to defects). When the fire extinguisher canister enters the inspection station, the system simultaneously activates a controllable ring array halogen lamp group to emit heat flux (the thermal excitation source activates the microscopic heat conduction on the canister surface) and triggers an infrared thermal imager to record the temperature data of the fire extinguisher canister surface at a fixed acquisition frequency, forming a time-sequential sequence of thermal imaging temperature data changes (this sequence records the dynamic process of abnormal heat conduction on the canister surface caused by defects). Finally, the time-sequential sequence of thermal imaging temperature data changes is integrated (the transient temperature characteristics are enhanced through an inter-frame difference algorithm) to generate thermal imaging video stream data (this data stores spatiotemporal-temperature correlation information in the form of a video stream, where high frame rate thermal imaging sequences can capture heat flux distortion caused by defects such as microcracks and dents). This thermal imaging video stream data is input into the defect identification algorithm: when an abnormal high temperature or low temperature sequence is detected in a local area of ​​the canister surface (such as a persistent low temperature patch caused by material loss in the rounded corner area of ​​the end cap), the system automatically triggers the sorting mechanism to remove the defective canister, ensuring the surface integrity of the fire extinguisher pressure vessel.

[0036] The scheme described in step 101 above achieves coordinated control of thermal excitation and temperature data acquisition on the surface of the fire extinguisher tank. By integrating a ring array of halogen lamps with an infrared thermal imager, an innovative synchronous system for uniform heat flow application and temperature monitoring is constructed. This technology employs a fixed acquisition frequency and time series integration method to transform discrete temperature data into a continuous thermal video stream, providing a complete time-series data foundation for subsequent analysis. The innovative thermal excitation source layout scheme ensures uniform distribution of heat flow on the tank surface, significantly improving the reliability and consistency of temperature monitoring data and creating an ideal thermal excitation environment for defect detection.

[0037] 102. By modulating the frequency and power of the thermal excitation source, a measurable difference in thermal response is made between the defective area and the normal area on the surface of the fire extinguisher canister due to their different thermal conductivity.

[0038] Optionally, step 102 may specifically include the following steps: 1021. Set the basic frequency and basic power parameters of the thermal excitation source, and adjust the frequency of the thermal excitation source from the basic frequency to the first frequency pulse mode, while increasing the power from the basic power to the first power level.

[0039] 1022. At the first power level, collect thermal imaging temperature data of the surface of the fire extinguisher tank and record the instantaneous change characteristics of the thermal imaging temperature data.

[0040] 1023. Restore the frequency of the thermal excitation source from the first frequency pulse mode to the fundamental frequency, and at the same time reduce the power from the first power level to the fundamental power.

[0041] 1024. Acquire thermal imaging temperature data of the surface of the fire extinguisher tank under the basic frequency and basic power modes, and record the steady-state characteristics of the thermal imaging temperature data.

[0042] 1025. Compare the instantaneous change characteristics with the steady-state characteristics to generate thermal response differences.

[0043] In the above scheme, the defective area refers to the surface area with defects. The normal area refers to the surface area without defects. Thermal conductivity refers to the relevant thermal conductivity characteristics of the material. Thermal response difference refers to the difference in thermal response. Fundamental frequency refers to the basic excitation frequency. Fundamental power parameters refer to the basic power parameters. First frequency pulse mode refers to the first frequency pulse mode. First power level refers to the first power level. Instantaneous change characteristics refer to instantaneous change characteristics. Steady-state characteristics refer to stable state characteristics.

[0044] In this embodiment, firstly, step 1021 sets the fundamental frequency and fundamental power parameters of the thermal excitation source, and adjusts the frequency of the thermal excitation source from the fundamental frequency to a first frequency pulse mode, while simultaneously increasing the power from the fundamental power to a first power level. This process, through the coordinated operation of the frequency modulation algorithm and the power drive circuit in the control system, first initializes the thermal excitation source to a stable fundamental operating state; then, pulse width modulation technology is used to switch the output frequency of the thermal excitation source to a pre-set first frequency pulse mode, which typically contains high-frequency pulse trains at specific intervals; simultaneously, by adjusting the driving voltage or current of the power device, the output power is gradually increased from the fundamental power level to a higher first power level, thereby applying a high-intensity thermal excitation condition with significant frequency characteristics to the surface of the fire extinguisher tank.

[0045] Subsequently, in step 1022, thermal imaging temperature data of the fire extinguisher tank surface is acquired at the first power level, and the instantaneous change characteristics of the thermal imaging temperature data are recorded. The infrared thermal imager synchronously captures the thermal radiation signal of the tank surface under high-temperature pulse excitation at a high frame rate and converts it into temperature matrix data; the signal processing algorithm analyzes the dynamic changes of the temperature data and extracts key indicators including the heating slope, peak response time, and local temperature gradient. These indicators together constitute the instantaneous change characteristics reflecting the transient behavior of heat conduction on the material surface.

[0046] Next, in step 1023, the frequency of the thermal excitation source is restored from the first frequency pulse mode to the fundamental frequency, while the power is reduced from the first power level to the fundamental power. The system gradually reduces the output of the power driver through the inverse modulation process, so that the power level of the thermal excitation source smoothly falls back to the initial fundamental power; at the same time, the frequency modulator is reconfigured to the fundamental frequency output mode, eliminating high-frequency pulse components and restoring the stable and continuous radiation state of the thermal excitation source, providing conditions for subsequent steady-state measurements.

[0047] Then, in step 1024, thermal imaging temperature data of the fire extinguisher tank surface is acquired under the basic frequency and basic power modes, and the steady-state characteristics of the thermal imaging temperature data are recorded. The thermal imager continuously monitors the heat distribution on the tank surface under constant thermal excitation and acquires stable data after the temperature reaches equilibrium. The feature extraction algorithm calculates parameters such as average temperature value, temperature distribution uniformity, and spatial hot spot pattern to form steady-state characteristics that characterize the inherent thermal conductivity of the material.

[0048] Finally, step 1025 compares the instantaneous change features with the steady-state features to generate thermal response differences. The data processing unit performs difference calculations or ratio analysis on the dynamic parameters (such as peak time and gradient value) in the instantaneous change features and the static parameters (such as average temperature and uniformity) in the steady-state features; through feature fusion algorithms, it highlights the differences in thermal behavior between normal areas and defective areas caused by different thermal conductivity, and finally generates a quantified thermal response difference matrix, which is used to identify the spatial location and severity of potential defects.

[0049] In practical applications, in the automatic sorting of surface defects in fire extinguisher canisters, operators place the canisters to be inspected on the automatic sorting line. First, they set the base frequency and power parameters of the thermal excitation source according to material and process requirements. Then, the frequency of the thermal excitation source is adjusted to the first-frequency pulse mode, and its power is increased to the first power level. In this mode, the canister surface is rapidly scanned and thermal imaging temperature data is collected, recording its instantaneous changes (e.g., areas suspected of corrosion exhibit abnormal heating rates due to differences in thermal conductivity). Next, the frequency of the thermal excitation source is restored to the base frequency, and the power is reduced to the base power. Thermal imaging temperature data of the canister surface is collected again, and its steady-state characteristics are recorded (e.g., normal areas have uniform temperature distribution, while defective areas show localized heat accumulation). Finally, by comparing the instantaneous changes and steady-state characteristics, a clear thermal response difference is generated. Based on this difference, the system automatically identifies and sorts out fire extinguisher canisters with defects, ensuring that product quality meets safety standards.

[0050] The scheme described in step 102 above achieves defect response feature extraction based on thermal excitation parameter modulation. An innovative method for analyzing the thermal response differences between defective and normal regions is designed through a dynamic adjustment mechanism of frequency and power. This technology employs a comparison strategy combining pulsed excitation and steady-state monitoring, effectively amplifying the differences in thermal response caused by different thermal conductivity characteristics. The innovative algorithm for comparing instantaneous changes and steady-state features significantly enhances defect features, providing high-contrast feature data for subsequent accurate identification and significantly improving the sensitivity of defect detection.

[0051] 103. Perform time-series thermal imaging analysis on the thermal video stream data to obtain the temperature decay curve and phase map in the thermal video stream data as time-series temperature data, and convert the time-series temperature data into frequency domain features through Fourier transform.

[0052] Optionally, step 103 may specifically include the following steps: 1031. Extract the temperature time series of each spatial location point from the thermal imaging video stream data, wherein the temperature time series includes the temperature measurement value at each time point.

[0053] 1032. Perform exponential decay curve fitting on the temperature time series to generate a temperature decay curve describing the temperature change over time.

[0054] 1033. Calculate the phase difference between adjacent time points in the temperature time series, and connect the phase differences in chronological order to form a phase diagram.

[0055] 1034. Integrate the temperature decay curve and the phase diagram into time-series temperature data, and obtain the change characteristics of the time-series temperature data.

[0056] 1035. The change features are converted into frequency domain amplitude feature values ​​and frequency domain phase feature values, and the frequency domain amplitude feature values ​​and frequency domain phase feature values ​​are combined to form frequency domain features.

[0057] In the above scheme, time-series thermal imaging analysis refers to the time-series thermal imaging analysis method. Temperature decay curve refers to the curve of temperature decay. Phase diagram refers to the phase graph. Time-series temperature data refers to time-series temperature data. Fourier transform refers to the transformation method from the time domain to the frequency domain. Frequency domain characteristics refer to the characteristics in the frequency domain. Temperature time series refers to the time series of temperature. Exponential decay curve fitting processing refers to the fitting processing of exponential decay curves. Phase difference refers to the phase difference. Variation characteristics refer to the characteristics of variation. Frequency domain amplitude characteristic values ​​refer to the characteristic values ​​of frequency domain amplitude. Frequency domain phase characteristic values ​​refer to the characteristic values ​​of frequency domain phase.

[0058] In this embodiment, firstly, step 1031 extracts the temperature time series of each spatial location point from the thermal video stream data. This temperature time series contains the temperature measurement values ​​at each time point. This process reads the thermal video stream frame by frame using a video frame parsing algorithm, arranging the temperature values ​​of each pixel in chronological order. For each spatial location point in the video, the system extracts and combines its temperature measurement values ​​from all frames into a one-dimensional array sorted by timestamp, thus forming a temperature time series reflecting the continuous change of temperature at that point over time, providing basic data for subsequent time series analysis.

[0059] Subsequently, step 1032 involves performing exponential decay curve fitting on the temperature time series to generate a temperature decay curve describing the temperature change over time. The system employs a nonlinear least squares method or optimization algorithm to fit the temperature time series to the exponential decay model. This model iteratively adjusts the initial temperature and decay rate parameters to ensure the curve optimally matches the actual temperature decrease trend, ultimately generating a smooth mathematical curve that quantitatively describes the temperature decay rate and steady state over time.

[0060] Next, step 1033 calculates the phase difference between adjacent time points in the temperature time series and connects the phase differences in chronological order to form a phase diagram. The system uses phase calculation algorithms (such as methods based on Hilbert transform or trigonometric functions) to calculate the phase shift value of temperature changes at adjacent time points; these phase difference values ​​are concatenated through a time axis to form a phase change sequence, and then the sequence is converted into a visual graphic through two-dimensional or three-dimensional mapping, where color or height represents the magnitude of the phase difference, thus forming a phase diagram that intuitively displays the temporal characteristics of temperature changes.

[0061] Then, in step 1034, the temperature decay curve and the phase diagram are integrated into time-series temperature data, and the variation characteristics of the time-series temperature data are obtained. The data fusion algorithm integrates the parameterized representation of the temperature decay curve (such as decay constant and initial temperature) with the matrix data of the phase diagram to form multidimensional time-series temperature data containing information in both the time domain and the phase domain. The feature extraction module then calculates key variation characteristics from this data, such as the temperature decrease slope, phase fluctuation amplitude, and periodic patterns, to comprehensively characterize the dynamic properties of thermal behavior.

[0062] Finally, in step 1035, the variation characteristics are converted into frequency domain amplitude and phase feature values, and these feature values ​​are combined to form frequency domain features. The system applies Fast Fourier Transform to convert the time-domain variation characteristics into a frequency-domain signal, decomposing it into an amplitude spectrum and a phase spectrum. The intensity values ​​of the main frequency components are extracted from the amplitude spectrum as frequency-domain amplitude feature values, and the phase angles of key frequencies are extracted from the phase spectrum as frequency-domain phase feature values. Finally, the two are combined into a composite frequency-domain feature vector through feature concatenation or weighted fusion to comprehensively describe the frequency distribution and phase characteristics of temperature changes.

[0063] In practical applications, such as the automated sorting of fire extinguisher canisters for surface defects, time-series thermal imaging analysis can be used to detect minute defects on the canister surface. First, fire extinguisher canisters passing at a constant speed on the production line are uniformly heated. Then, thermal imagers capture thermal video stream data of their cooling process. For each spatial location in the video, a temperature time series is extracted, containing the temperature measurements at each time point. By fitting these temperature time series to exponential decay curves, a temperature decay curve describing the temperature change over time is generated. The curve for intact canisters is usually smooth and regular, while defective areas exhibit different decay patterns due to abnormal thermal conductivity. Simultaneously, the phase difference between adjacent time points in the temperature time series is calculated, and these phase differences are connected chronologically to form a phase map, which reflects the temporal consistency of heat wave transmission. After integrating the temperature decay curve and the phase map into time-series temperature data, its variation characteristics are obtained. Finally, these changing features are converted into frequency domain amplitude feature values ​​and frequency domain phase feature values ​​through Fourier transform, and the two are combined to form frequency domain features. Complete cans exhibit concentrated energy distribution in the frequency domain, while cans with defects such as paint peeling, dents or uneven internal materials will show abnormal frequency domain features, thus being identified and rejected by the automatic sorting system.

[0064] The scheme described in step 103 above achieves multi-dimensional feature extraction and frequency domain transformation of temperature time-series data. Through temperature decay curve fitting and phase analysis, an innovative dual feature representation system in the time and frequency domains is constructed. This technique employs a combination of an exponential decay model and phase difference calculation to comprehensively describe the dynamic characteristics of temperature changes. The innovative Fourier transform processing converts time-series features into frequency domain amplitude and phase indices, providing richer feature dimensions for defect identification and significantly improving the depth and breadth of feature analysis.

[0065] 104. Analyze the thermal response differences using the frequency domain characteristics to identify the thermal signal difference regions between defective and normal regions.

[0066] Optionally, step 104 may specifically include the following steps: 1041. Extract the frequency domain amplitude feature value and frequency domain phase feature value of each spatial location point from the frequency domain features.

[0067] 1042. Obtain the thermal response difference amount corresponding to the spatial location point in the thermal response difference, wherein the thermal response difference amount describes the degree of difference in thermal conductivity between the defective area and the normal area.

[0068] 1043. Combine the frequency domain amplitude feature value with the thermal response difference to generate a comprehensive signal strength value, and compare the comprehensive signal strength value with a preset normal area signal strength threshold. When the comprehensive signal strength value exceeds the normal area signal strength threshold, generate an abnormal marker signal for the spatial location point.

[0069] 1044. Integrate the abnormal marker signals of all spatial locations and arrange them in order of spatial coordinates to form a thermal signal difference region that characterizes the difference between the defective area and the normal area.

[0070] In the above scheme, the thermal signal difference region refers to the region where the thermal signal differs. The comprehensive signal strength value refers to the overall signal strength value. The normal region signal strength threshold refers to the critical signal strength value in the normal region. The abnormal marker signal refers to the signal that marks an abnormality. The degree of difference in thermal conductivity refers to the degree of difference in thermal characteristics.

[0071] In this embodiment, firstly, step 1031 extracts the temperature time series of each spatial location point from the thermal video stream data. This temperature time series contains the temperature measurement values ​​at each time point. This process reads the thermal video stream frame by frame using a video frame parsing algorithm, arranging the temperature values ​​of each pixel in chronological order. For each spatial location point in the video, the system extracts and combines its temperature measurement values ​​from all frames into a one-dimensional array sorted by timestamp, thus forming a temperature time series reflecting the continuous change of temperature at that point over time, providing basic data for subsequent time series analysis.

[0072] Subsequently, step 1032 involves performing exponential decay curve fitting on the temperature time series to generate a temperature decay curve describing the temperature change over time. The system employs a nonlinear least squares method or optimization algorithm to fit the temperature time series to the exponential decay model. This model iteratively adjusts the initial temperature and decay rate parameters to ensure the curve optimally matches the actual temperature decrease trend, ultimately generating a smooth mathematical curve that quantitatively describes the temperature decay rate and steady state over time.

[0073] Next, step 1033 calculates the phase difference between adjacent time points in the temperature time series and connects the phase differences in chronological order to form a phase diagram. The system uses phase calculation algorithms (such as methods based on Hilbert transform or trigonometric functions) to calculate the phase shift value of temperature changes at adjacent time points; these phase difference values ​​are concatenated through a time axis to form a phase change sequence, and then the sequence is converted into a visual graphic through two-dimensional or three-dimensional mapping, where color or height represents the magnitude of the phase difference, thus forming a phase diagram that intuitively displays the temporal characteristics of temperature changes.

[0074] Then, in step 1034, the temperature decay curve and the phase diagram are integrated into time-series temperature data, and the variation characteristics of the time-series temperature data are obtained. The data fusion algorithm integrates the parameterized representation of the temperature decay curve with the matrix data of the phase diagram to form multidimensional time-series temperature data containing information in both the time domain and the phase domain. The feature extraction module then calculates key variation characteristics from this data, such as the temperature decrease slope, phase fluctuation amplitude, and periodic patterns, to comprehensively characterize the dynamic properties of thermal behavior.

[0075] Finally, in step 1035, the variation characteristics are converted into frequency domain amplitude and phase feature values, and these feature values ​​are combined to form frequency domain features. The system applies Fast Fourier Transform to convert the time-domain variation characteristics into a frequency-domain signal, decomposing it into an amplitude spectrum and a phase spectrum. The intensity values ​​of the main frequency components are extracted from the amplitude spectrum as frequency-domain amplitude feature values, and the phase angles of key frequencies are extracted from the phase spectrum as frequency-domain phase feature values. Finally, the two are combined into a composite frequency-domain feature vector through feature concatenation or weighted fusion to comprehensively describe the frequency distribution and phase characteristics of temperature changes.

[0076] In practical applications, specifically in the automatic sorting system for surface defects on fire extinguisher canisters, the process begins with uniform thermal excitation of the canisters on the production line (e.g., using short-time infrared heating lamps), followed by the acquisition of thermal image sequences of the canister surface during cooling using an infrared thermal imager. This sequence is then subjected to a Fourier transform to extract the frequency domain amplitude and phase features of each pixel (i.e., spatial location). These frequency domain features reflect the differences in thermal diffusion characteristics between different areas. Next, the system calculates the thermal response difference at each spatial location, quantifying the degree of difference in thermal behavior caused by the different thermal conductivity characteristics between potential defects (such as internal bubbles, material delamination, or uneven thickness) and normal areas. To further highlight anomalies, the frequency domain amplitude feature and thermal response difference at each point are weighted and fused to generate a comprehensive signal strength value. If this strength value exceeds a preset normal area signal strength threshold based on normal canister samples, the system generates an anomaly marker signal for that location. Finally, these abnormal marker signals from all spatial locations are integrated and arranged according to their two-dimensional spatial coordinates to form a clear, binary thermal signal difference map. This image intuitively identifies the boundary between defective and normal areas and directly drives the subsequent automatic sorting mechanism to remove unqualified cans from the production line.

[0077] The scheme described in step 104 above achieves accurate identification of defective areas and localization of thermal signal differences. Based on the fusion analysis of frequency domain features and thermal response differences, an innovative comprehensive signal strength evaluation model is constructed. This technology employs a threshold comparison and anomaly marking mechanism to achieve automatic identification and localization of defective areas. An innovative spatial coordinate integration method transforms discrete anomaly points into continuous thermal signal difference regions, providing a clear target area for subsequent segmentation processing and significantly improving the accuracy and efficiency of defect localization.

[0078] 105. The thermal signal difference area is segmented, and the segmented area is classified into defect types through morphological processing to output a sorting signal, and automatic sorting is performed according to the sorting signal.

[0079] Optionally, step 105 may specifically include the following steps: 1051. Obtain the thermal signal difference value of the thermal signal difference region, and mark the region where the thermal signal difference value exceeds the preset segmentation threshold as the foreground region, and the region where the thermal signal difference value does not exceed the preset segmentation threshold as the background region.

[0080] 1052. Connect adjacent foreground regions and fill the internal voids of the foreground regions to form connected regions.

[0081] 1053. Extract the morphological feature parameters of the connected regions, match the morphological feature parameters with preset defect type features, and determine the defect type identifier corresponding to each connected region based on the matching result.

[0082] Step 1053 may specifically include the following processes: measuring the area value of each connected region, and simultaneously calculating the length-to-width ratio of the bounding rectangle of the connected region as an aspect ratio feature value; calculating the ratio of the number of boundary pixels of the connected region to the square of the perimeter of the connected region as a boundary complexity feature value; combining the area value, aspect ratio feature value, and boundary complexity feature value into morphological feature parameters; comparing the morphological feature parameters with a preset defect type feature threshold to generate a successful match signal; and determining the defect type identifier of the connected region based on the defect type feature threshold corresponding to the successful match signal.

[0083] 1054. Generate a sorting instruction signal based on the defect type identifier, and send the sorting instruction signal to the automatic sorting device to perform the sorting operation.

[0084] In the above scheme, segmentation refers to the image segmentation operation. Morphological processing refers to morphological image processing. Defect type classification refers to the classification of defect types. Sorting signal refers to the signal controlling sorting. Automatic sorting refers to the automatic sorting operation. Thermal signal difference value refers to the numerical value of thermal signal difference. Preset segmentation threshold refers to the preset critical value for segmentation. Foreground region refers to the foreground region of the image. Background region refers to the background region of the image. Connected region refers to a connected image region. Morphological feature parameters refer to morphological feature parameters. Defect type identifier refers to the identifier of defect type. Region area value refers to the area value of the region. Circumscribed rectangle refers to the bounding rectangle. Aspect ratio feature value refers to the feature value of the aspect ratio. Number of boundary pixels refers to the number of boundary pixels. Perimeter square refers to the square value of the perimeter. Boundary complexity feature value refers to the feature value of the boundary complexity. Defect type feature threshold refers to the feature critical value of the defect type. Match success signal refers to the signal of successful matching. Sorting instruction signal refers to the instruction signal controlling sorting. Automatic sorting device refers to the equipment that automatically performs sorting. Sorting operation refers to the specific sorting operation.

[0085] In this embodiment, firstly, the thermal signal difference value of the thermal signal difference region is obtained through step 1051, and the region whose thermal signal difference value exceeds the preset segmentation threshold is marked as the foreground region, and the region whose thermal signal difference value does not exceed the preset segmentation threshold is marked as the background region. This process uses an image segmentation algorithm to scan the difference intensity value of each pixel in the thermal signal difference region, and compares the thermal signal difference value of each pixel with the preset segmentation threshold; when the difference value of a pixel exceeds the threshold, it is marked as the foreground region (suspected defect region), while pixels with a difference value lower than or equal to the threshold are classified as the background region (normal region), thereby generating a binary segmentation image, in which the white area represents the foreground and the black area represents the background, completing the initial separation of the defect region.

[0086] Subsequently, in step 1052, adjacent foreground regions are connected and their internal holes are filled to form connected regions. The system uses a morphological closing operation algorithm (dilation followed by erosion) to process the binarized image: the dilation operation connects adjacent foreground region pixels into patches, eliminating tiny gaps and forming continuous blocks; the erosion operation then adjusts the region boundaries to make them smoother; at the same time, a hole-filling algorithm is used to scan the internal pixels of each foreground block. If background pixels surrounded by the foreground (i.e., holes) are found, they are converted into foreground pixels to ensure that each connected region is filled and free of holes, ultimately forming several connected regions with complete boundaries and continuous internal structures, laying the foundation for subsequent feature extraction.

[0087] Next, morphological feature parameters of the connected regions are extracted in step 1053, and these morphological feature parameters are matched with preset defect type features. Based on the matching results, the defect type identifier corresponding to each connected region is determined. This step first measures the total number of pixels in each connected region as the region area value, and simultaneously calculates the ratio of the length to the width of its minimum bounding rectangle as the aspect ratio feature value. Then, the contour pixel sequence of the connected region is obtained through a boundary tracking algorithm, and the ratio of the number of contour pixels to the square of the perimeter is calculated as the boundary complexity feature value. These three feature values ​​are combined to form the morphological feature parameters. The system then compares these parameters one by one with preset defect type feature thresholds (such as area range, aspect ratio range, and complexity range). If the parameter falls within the threshold range of a certain type of defect, a matching success signal is generated. Finally, based on the defect category (such as cracks, pores, inclusions, etc.) corresponding to the matching signal, a unique defect type identifier is assigned to each connected region, completing the defect classification.

[0088] Finally, in step 1054, a sorting instruction signal is generated based on the defect type identifier, and the sorting instruction signal is sent to the automatic sorting device to perform the sorting operation. The control system converts the defect type identifier into a predefined sorting instruction code (such as different sorting actions corresponding to different defect types), and sends the instruction signal to the automatic sorting device through an industrial communication protocol; after receiving the signal, the sorting device drives the actuator (such as a robotic arm or pneumatic push rod) to sort the corresponding workpiece to the designated area (such as a scrap box or rework area), realizing automated sorting based on defect type.

[0089] In practical applications, this is demonstrated in the implementation of an automatic defect sorting system for fire extinguisher canister surfaces. The system first acquires an image of the thermal signal difference region generated by frequency domain analysis and calculates the thermal signal difference value for each pixel. This value is compared with a preset segmentation threshold; regions exceeding the threshold are marked as foreground regions, while those below are marked as background regions, thus initially separating potential defects. Subsequently, the system performs morphological processing on the foreground regions, connecting adjacent foreground pixels and filling internal voids to form complete connected regions, accurately representing the actual shape and extent of the defects. Next, morphological feature parameters for each connected region are extracted, including region area, aspect ratio (calculated using the circumscribed rectangle size), and boundary complexity (based on the relationship between boundary pixels and perimeter). These parameters are matched with preset defect type feature thresholds (such as the geometric features of typical defects like scratches, dents, or bubbles), generating a successful match signal, and thus determining the defect type identifier for each connected region. Finally, the system generates a sorting instruction signal based on the defect type identifier and sends it to an automatic sorting device (such as a robotic arm or sorting conveyor belt) to drive it to perform sorting operations and classify fire extinguisher canisters with different defect types into designated areas.

[0090] The solution described in step 105 above achieves intelligent classification and automated sorting of defect types. Through morphological feature extraction and classification matching, an innovative multi-parameter defect classification system is constructed. This technology employs a combination of regional connectivity and boundary complexity analysis to achieve accurate differentiation between different types of defects. An innovative sorting instruction generation mechanism directly converts detection results into execution instructions, realizing a complete closed loop from detection to sorting. This technical solution, combining morphological features with automated sorting, significantly improves the automation level and processing efficiency of product quality control.

[0091] The following are specific examples for steps 101 to 105: In the practical application of the automatic sorting system for surface defects in fire extinguisher canisters, the system first transports the fire extinguisher canisters to the inspection station via a production line conveyor belt. At this point, a controllable ring array of halogen lamps above the station applies a uniform heat flow to the canister surface. Simultaneously, an infrared thermal imager records the temperature data changes of the canister surface at a fixed acquisition frequency, generating a continuous thermal video stream. By modulating the frequency and power of the thermal excitation source, the system effectively distinguishes the differences in thermal response caused by the different thermal conductivity of defective and normal areas. Subsequently, time-series thermal image analysis is performed on the thermal video stream data, extracting the temperature decay curve and phase map of each spatial location as time-series temperature data. This data is then converted into frequency domain features containing frequency domain amplitude and phase characteristics using Fourier transform. Utilizing these frequency domain features and combining them with the thermal response difference, the system generates a comprehensive signal strength value for each location. This value is compared with a preset normal area signal strength threshold to generate an anomaly marker signal. Finally, the signals from all spatial points are integrated to form a thermal signal difference region. After segmenting the region, regions exceeding the preset segmentation threshold are marked as foreground regions. Adjacent foreground regions are connected and internal voids are filled to form connected regions. Then, the morphological feature parameters of each connected region (including region area value, aspect ratio feature value, and boundary complexity feature value) are extracted and matched with the preset defect type features to determine the defect type identifier. Finally, a sorting instruction signal is generated based on the identifier to drive the automatic sorting device to perform sorting operations.

[0092] Figure 2 This application provides a schematic diagram of an automatic sorting system for surface defects of fire extinguisher canisters based on machine vision, as shown in the embodiment of the present application. Figure 2 As shown, the system includes: The acquisition module 21 is used to apply a uniform heat flow to the surface of the fire extinguisher tank using an infrared thermal imager and a controllable ring array halogen lamp group as thermal excitation sources, and at the same time acquire the change sequence of thermal imaging temperature data of the surface of the fire extinguisher tank to generate thermal imaging video stream data. The differentiation module 22 is used to differentiate between the defective areas and normal areas on the surface of the fire extinguisher canister by modulating the frequency and power of the thermal excitation source to produce measurable differences in thermal response due to different thermal conductivity. The conversion module 23 is used to perform time-series thermal image analysis on the thermal video stream data to obtain the temperature decay curve and phase map in the thermal video stream data as time-series temperature data, and convert the time-series temperature data into frequency domain features through Fourier transform. The identification module 24 is used to analyze the thermal response differences using the frequency domain features to identify the thermal signal difference areas between the defective areas and the normal areas. The output module 25 is used to segment the thermal signal difference area, classify the defect type of the segmented area through morphological processing, output a sorting signal, and perform automatic sorting according to the sorting signal.

[0093] Figure 2 The aforementioned automatic sorting system for surface defects of fire extinguisher canisters based on machine vision can perform... Figure 1 The implementation principle and technical effects of the automatic sorting method for surface defects of fire extinguisher canisters based on machine vision, as described in the illustrated embodiment, will not be repeated here. The specific methods by which each module and unit of the automatic sorting system for surface defects of fire extinguisher canisters based on machine vision in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0094] In one possible design, Figure 2 The machine vision-based automatic sorting system for surface defects of fire extinguisher canisters, as shown in the embodiment, can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0095] The processing component 32 is used for the above Figure 1 The embodiment describes an automatic sorting method for surface defects on fire extinguisher canisters based on machine vision.

[0096] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0097] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0098] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0099] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0100] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0101] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0102] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is an automatic sorting method for surface defects of fire extinguisher canisters based on machine vision.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] 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.

[0105] 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, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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. Such 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 this application.

Claims

1. A method for automatically sorting surface defects of fire extinguisher tank bodies based on machine vision, characterized by, The method comprises the following steps: Using an infrared thermal imager and a controllable ring array halogen lamp group as a heat excitation source to apply uniform heat flow to the surface of the fire extinguisher tank, and collecting the change sequence of the thermal imaging temperature data of the surface of the fire extinguisher tank to generate thermal image video stream data; By modulating the frequency and power of the heat excitation source, the measurable thermal response difference between the defect area and the normal area of the fire extinguisher tank surface due to the difference in thermal conductivity characteristics is distinguished; Performing time-series thermal image analysis on the thermal image video stream data to obtain the temperature decay curve and phase diagram in the thermal image video stream data as time-series temperature data, and converting the time-series temperature data into frequency domain features by Fourier transform; Using the frequency domain features to analyze the thermal response difference to identify the thermal signal difference area between the defect area and the normal area; Segmenting the thermal signal difference area and classifying the segmented area by morphological processing to output a sorting signal and automatically sorting according to the sorting signal.

2. The method of claim 1, wherein, Using an infrared thermal imager and a controllable ring array halogen lamp group as a heat excitation source to apply uniform heat flow to the surface of the fire extinguisher tank, and collecting the change sequence of the thermal imaging temperature data of the surface of the fire extinguisher tank to generate thermal image video stream data, comprising: A controllable ring array halogen lamp group is installed in a ring above the fire extinguisher tank conveying line, and the irradiation angle of the controllable ring array halogen lamp group covers the surface of the fire extinguisher tank and the heat flow is uniformly distributed; An infrared thermal imager is installed at the controllable ring array halogen lamp group, and the focal length of the infrared thermal imager is adjusted to make the field of view of the infrared thermal imager completely cover the surface area of the fire extinguisher tank being tested; When the fire extinguisher tank enters the detection station, the controllable ring array halogen lamp group is started to emit heat flow and the infrared thermal imager is triggered to record the temperature data of the surface of the fire extinguisher tank at a fixed collection frequency to form a change sequence of thermal imaging temperature data arranged in time sequence; The change sequence of the thermal imaging temperature data is integrated in time sequence to generate thermal image video stream data.

3. The method of claim 1, wherein, Segmenting the thermal signal difference area and classifying the segmented area by morphological processing to output a sorting signal and automatically sorting according to the sorting signal, comprising: Obtaining the thermal signal difference value of the thermal signal difference area, and marking the area with a thermal signal difference value exceeding a preset segmentation threshold as a foreground area, and marking the area without a thermal signal difference value exceeding the preset segmentation threshold as a background area; Connecting adjacent foreground areas and filling the internal cavities of the foreground areas to form connected regions; Extracting morphological feature parameters of the connected regions, matching the morphological feature parameters with preset defect type features, and determining the defect type identifier corresponding to each connected region according to the matching result; Generating a sorting instruction signal according to the defect type identifier, and sending the sorting instruction signal to an automatic sorting device to perform a sorting operation.

4. The method of claim 3, wherein, Extracting morphological feature parameters of the connected regions, matching the morphological feature parameters with preset defect type features, and determining the defect type identifier corresponding to each connected region according to the matching result, comprising: The area value of each connected region is measured, and the length-width ratio of the circumscribed rectangle of the connected region is calculated as a length-width ratio characteristic value; The ratio of the number of boundary pixels of the connected region to the square of the perimeter of the connected region is calculated as a boundary complexity characteristic value; The area value, length-width ratio characteristic value and boundary complexity characteristic value are combined as morphological feature parameters; The morphological feature parameters are compared with preset defect type feature thresholds to generate a matching success signal; The defect type identification of the connected region is determined according to the defect type feature threshold corresponding to the matching success signal.

5. The method as claimed in claim 1, wherein, The thermal response difference is analyzed by using the frequency domain feature to identify the thermal signal difference region between the defect region and the normal region, including: The frequency domain amplitude feature value and the frequency domain phase feature value of each spatial position point are extracted from the frequency domain feature; The thermal response difference quantity of the corresponding spatial position point in the thermal response difference is obtained, which describes the difference degree of the thermal conductivity characteristics between the defect region and the normal region; The frequency domain amplitude feature value and the thermal response difference quantity are combined to generate a comprehensive signal strength value, and the comprehensive signal strength value is compared with a preset normal region signal strength threshold, and when the comprehensive signal strength value exceeds the normal region signal strength threshold, an abnormal mark signal of the spatial position point is generated; All abnormal mark signals of the spatial position points are integrated and arranged in order of spatial coordinates to form a thermal signal difference region representing the difference between the defect region and the normal region.

6. The method as claimed in claim 1, wherein, The thermal image video stream data is analyzed to obtain the temperature decay curve and the phase diagram in the thermal image video stream data as time sequence temperature data, and the time sequence temperature data is converted into frequency domain features by Fourier transform, including: The temperature time series of each spatial position point is extracted from the thermal image video stream data, and the temperature time series includes the temperature measurement value at each time point; The temperature time series is subjected to exponential decay curve fitting processing to generate a temperature decay curve describing the temperature change rule over time; The phase difference between adjacent time points in the temperature time series is calculated, and the phase difference is connected in time sequence to form a phase diagram; The temperature decay curve and the phase diagram are integrated into time sequence temperature data, and the change characteristics of the time sequence temperature data are obtained; The change characteristics are converted into frequency domain amplitude feature values and frequency domain phase feature values, and the frequency domain amplitude feature values and the frequency domain phase feature values are combined to form frequency domain features.

7. The method as claimed in claim 1, wherein, The frequency and power of the thermal excitation source are modulated to distinguish the measurable thermal response difference between the defect region and the normal region on the surface of the fire extinguisher tank due to the difference in thermal conductivity, including: The base frequency and base power parameters of the thermal excitation source are set, the frequency of the thermal excitation source is adjusted from the base frequency to a first frequency pulse mode, and the power is increased from the base power to a first power level; The thermal imaging temperature data of the fire extinguisher tank surface is collected at the first power level, and the instantaneous change characteristics of the thermal imaging temperature data are recorded; restoring the frequency of the thermal excitation source from the first frequency pulse mode to a base frequency while reducing the power from a first power level to a base power; acquiring thermal imaging temperature data of the surface of the fire extinguisher tank body in the base frequency and base power mode, and recording steady state characteristics of the thermal imaging temperature data; comparing the transient change characteristics and the steady state characteristics to generate thermal response differences.

8. A machine vision based automatic sorting system for surface blemish of fire extinguisher tank, characterized in that, comprise: an acquisition module for applying uniform heat flow to the surface of the fire extinguisher tank body using an infrared thermal imager and a controllable annular array halogen lamp group as a thermal excitation source, while acquiring a change sequence of thermal imaging temperature data of the surface of the fire extinguisher tank body to generate thermal imaging video stream data; a distinguishing module for distinguishing between the defect area and the normal area of the surface of the fire extinguisher tank body by modulating the frequency and power of the thermal excitation source to produce measurable thermal response differences due to different thermal conductivity characteristics; a conversion module for performing time-series thermal imaging analysis on the thermal imaging video stream data to obtain temperature decay curves and phase diagrams in the thermal imaging video stream data as time-series temperature data, and converting the time-series temperature data into frequency domain characteristics through Fourier transform; an identification module for analyzing the thermal response differences using the frequency domain characteristics to identify thermal signal difference areas between the defect area and the normal area; an output module for segmenting the thermal signal difference areas and classifying the segmented areas by defect type through morphological processing to output sorting signals and automatically sort according to the sorting signals.

9. A computing device, comprising: comprise a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a machine vision-based automatic sorting method for surface defects of a fire extinguisher tank body according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that a computer program is stored, and when the computer program is executed by a computer, a machine vision-based automatic sorting method for surface defects of a fire extinguisher tank body according to any one of claims 1 to 7 is implemented.

Images