High-temperature special steel casting blank surface detection multispectral sensor system and method

Abstract

The invention discloses a multispectral sensor system and method for detecting the surface of a high-temperature special steel casting blank. According to the multispectral sensor system, a plurality of image sensor chips are adopted, and the chips perform synchronous imaging in different spectral sections, so that the multispectral image of the surface of the special steel casting blank is obtained. In order to process the multispectral images, a heterogeneous high-speed calculation processing unit based on an FPGA (Field Programmable Gate Array), a GPU (Graphic Processing Unit) and a CPU (Central Processing Unit) is adopted. The processing unit can realize synchronous acquisition and high-speed parallel calculation, so that the acquisition and processing requirements of high-dynamic and high-concurrency data volume are met. According to the method, high-quality surface images of the special steel casting blanks can be provided, and the defect detection requirements of the high-speed casting blanks can be met.

Description

Technical Field

[0001] The invention belongs to the field of steel high-temperature continuous casting surface quality detection, and relates to a multi-spectral sensor system and method for high-temperature special steel ingot surface detection. Background Art

[0002] Special steel ingots refer to steel ingots produced under high temperature conditions. They have the characteristics of high strength, high toughness, high hardness, high wear resistance, etc. They are widely used in aerospace, automobile manufacturing, machinery manufacturing and other fields. Since the production process of special steel ingots is complicated and defects are difficult to avoid, it is of great significance to detect defects on the surface of special steel ingots.

[0003] The surface defect of characteristic steel is a common defect, which not only affects the appearance of the steel but also easily causes rust and other phenomena on the steel itself, reducing the use function of the steel. In severe cases, it may lead to the scrapping of the steel and cause unnecessary economic losses.

[0004] The cracks of high-temperature castings mainly include surface longitudinal cracks, surface transverse cracks, network cracks, surface slag, bubbles and pores.

[0005] With the development of special steel casting technology, higher requirements are put forward for high-temperature special steel ingot surface detection technology. At present, domestic and foreign scholars are actively studying new high-temperature special steel ingot surface detection technology to meet the needs of special steel casting production.

[0006] The following are some high temperature special steel ingot surface inspection technologies:

[0007] 1. Laser detection: Laser detection uses the reflection characteristics of the laser beam on the surface of the special steel ingot to detect the surface of the special steel ingot. Laser detection has high detection accuracy and sensitivity, but its disadvantages are that the energy of the laser beam is high and the detection cost is high.

[0008] 2. Electromagnetic wave detection: Electromagnetic wave detection uses the propagation characteristics of electromagnetic waves on the surface of special steel ingots to detect the surface of special steel ingots. Electromagnetic wave detection has high detection accuracy and sensitivity, but its disadvantages are that the energy of electromagnetic waves is high and the detection cost is high.

[0009] 3. Spectral detection: Spectral detection uses the spectral characteristics of the surface of special steel ingots to detect the surface of special steel ingots. Spectral detection has high detection accuracy and sensitivity, but its disadvantage is that the spectral detection equipment is relatively complex and the detection cost is high.

[0010] Due to the high temperature and harsh environment of the continuous casting process, it is impossible to observe the surface of the ingot with the naked eye, and the radiation light and iron oxide scale on the surface of the high-temperature ingot have a great impact on the imaging and image recognition of the machine vision method. Therefore, the acquired images are often disturbed by factors such as ambient light, high humidity and aerosol, resulting in poor image quality and inaccurate detection results. The crack size of high-temperature ingots is generally around 100um, requiring a high resolution for the imaging system. In addition, due to the high speed and continuity of the continuous casting process, extremely high requirements are placed on the acquisition and processing of image data, and existing processing units often cannot meet these requirements.

[0011] Therefore, how to design a multispectral sensor that can detect surface defects of special steel ingots in real time and accurately is an important research topic at present. Summary of the invention

[0012] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art and to provide a multi-spectral sensor system and method for detecting the surface of a high-temperature special steel ingot. The multi-spectral sensor system can detect defects on the surface of the special steel ingot in real time and accurately.

[0013] The technical solution of the present invention is: in order to realize real-time and accurate detection of defects on the surface of special steel ingots, the present invention provides a multispectral sensor system for high-temperature special steel ingot surface detection. The system includes an image acquisition unit, an FPGA module, a GPU module, a PCI E bus, a CPU module, a DDR module and a transmission module. Among them, the image acquisition unit acquires a multispectral image of the surface of the steel ingot and sends it to the FPGA module. The FPGA module processes the multispectral image of the surface of the steel ingot and sends the processed multispectral image to the GPU module. The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral image and sends it to the DDR module. The CPU module reads the multispectral image from the DDR module and sends it to the transmission module. The transmission module sends the multispectral image to an external server.

[0014] The design method of the present invention mainly includes the following:

[0015] The present invention discloses a multi-spectral sensor system for detecting the surface of a high-temperature special steel ingot, comprising: an image acquisition unit, an FPGA module, a GPU module, a PCIE bus, a CPU module, a DDR module and a transmission module; wherein:

[0016] An image acquisition unit collects multi-spectral images of the steel ingot surface and sends them to the FPGA module;

[0017] The FPGA module processes the multispectral image of the steel ingot surface and sends the processed multispectral image to the GPU module;

[0018] The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module;

[0019] The CPU module reads the multispectral image from the DDR module and sends it to the transmission module;

[0020] Transmission module,sends multispectral images to external servers.

[0021] Furthermore, in the above-mentioned sensor system, the image acquisition unit includes a multispectral optical component and an imaging sensitive unit; the multispectral optical component is used to acquire a multispectral image of the steel ingot surface; the imaging sensitive unit is used to convert the multispectral image of the steel ingot surface into an electrical signal.

[0022] Furthermore, in the above-mentioned sensor system, the multi-spectral optical component includes a mounting plate, an R-band lens assembly, a G-band lens assembly, a B-band lens assembly and a far-infrared band lens assembly; wherein the R-band lens assembly, the G-band lens assembly, the B-band lens assembly and the far-infrared band lens assembly are evenly distributed in a square shape, and the angles between the optical axes of the R-band lens assembly, the G-band lens assembly, the B-band lens assembly and the far-infrared band lens assembly and the normal of the mounting plate are all 15 to 25 degrees.

[0023] Furthermore, in the above sensor system, the R-band lens assembly is provided with a filter with a spectral range of 645 to 655 nm; the G-band lens assembly is provided with a filter with a spectral range of 525 to 535 nm; and the B-band lens assembly is provided with a filter with a spectral range of 445 to 455 nm.

[0024] Furthermore, in the above-mentioned sensor system, the FPGA module includes three visible light image processing modules, an infrared image processing module and a synchronization module; wherein the synchronization module sends a trigger signal to the three visible light image processing modules and the infrared image processing module; the three visible light image processing modules and the infrared image processing modules perform image acquisition and image processing according to the trigger signal, and send the processed multispectral image to the GPU module.

[0025] Furthermore, in the above-mentioned sensor system, the visible light image processing module includes a visible light image acquisition module, a bad pixel correction module, a visible light filtering processing module, a grayscale conversion module, a visible light correction module and an image enhancement module; the image is sequentially acquired, bad pixel corrected, filtered, grayscale converted, image corrected and enhanced to generate a visible light image, which is sent to the GPU module.

[0026] Furthermore, in the above sensor system, the infrared image processing module includes: an infrared image acquisition module, a non-uniformity correction module, an infrared filter processing module, a DDE processing module and a temperature calculation module; wherein,

[0027] The non-uniformity correction module performs shutter correction on the data collected by the infrared image acquisition module, and performs KB correction on the shutter-corrected data to obtain a uniform RAW image; the infrared filtering processing module performs filtering on the RAW image and sends it to the DDE processing module and the temperature calculation module; the DDE processing module performs image enhancement on the RAW image after filtering, obtains an enhanced image, and sends it to the GPU module; the temperature calculation module performs temperature calibration and calculation on the RAW image after filtering, obtains temperature data, and sends it to the GPU module.

[0028] Furthermore, in the above sensor system, the DDE processing module includes a high- and low-frequency separation processing module, an adaptive local enhancement module, and an infrared correction module; wherein,

[0029] The high- and low-frequency separation processing module performs Gaussian filtering on the filtered RAW image to obtain a low-frequency image, and subtracts the low-frequency image from the filtered RAW image to obtain a high-frequency image; the adaptive local enhancement module performs adaptive local enhancement on the low-frequency image and the high-frequency image, fuses the enhanced high-frequency image and the low-frequency image and converts them into a grayscale image; the infrared correction module corrects the grayscale image and sends the corrected grayscale image to the GPU module.

[0030] A method for detecting a high-temperature special steel ingot surface using a multi-spectral sensor system comprises:

[0031] The image acquisition unit acquires multi-spectral images of the steel ingot surface and sends them to the FPGA module;

[0032] The FPGA module processes the multispectral image of the steel ingot surface and sends the processed multispectral image to the GPU module;

[0033] The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module;

[0034] The CPU module reads the multispectral image from the DDR module and sends it to the transmission module;

[0035] The transmission module sends the multispectral images to an external server.

[0036] The advantages of the present invention compared with the prior art are:

[0037] 1. The present invention uses multispectral imaging to simultaneously obtain three band images and thermal imaging images within the visible light range. Using different spectral imaging can enhance feature recognition, improve image contrast, reduce environmental interference, provide more information, and detect temperature distribution through thermal imaging. These are all helpful to improve the accuracy and efficiency of surface defect detection of special steel ingots.

[0038] 2. The present invention uses large array multi-spectral real-time direct imaging, which can provide real-time higher image resolution, making the details in the image clearer, helping to more accurately detect and identify surface defects and meet real-time detection requirements.

[0039] 3. The multispectral image of the present invention uses heterogeneous computing units based on FPGA modules, GPU modules and CPU modules to achieve high-speed parallel computing, clearly divide tasks, and use high-speed PCI E data bus for fast data transmission, thereby improving data processing speed and efficiency and meeting the image processing needs of large data volumes.

[0040] 4. The high-temperature special steel ingot surface detection multi-spectral sensor system of the present invention can detect surface defects of special steel ingots in real time and accurately under harsh on-site environments, thereby improving production efficiency and product quality. In addition, the design of the present invention is also characterized by high speed, high efficiency, stability and scalability, enabling it to adapt to various complex industrial environments and needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a block diagram of the system composition of the present invention;

[0042] Figure 2 It is a block diagram of the image acquisition unit of the multi-spectral sensor system of the present invention;

[0043] Figure 3 It is a block diagram of the multi-spectral sensor FPGA module of the present invention;

[0044] Figure 4 It is the multi-spectral 445-455nm band imaging diagram of the present invention;

[0045] Figure 5 It is the multi-spectral 525-535nm band imaging diagram of the present invention;

[0046] Figure 6 It is the multi-spectral 645-655nm band imaging diagram of the present invention;

[0047] Figure 7 This is the multi-spectral thermal imaging image of the present invention. DETAILED DESCRIPTION

[0048] The working principle and working process of the present invention will be further explained and illustrated below in conjunction with the accompanying drawings.

[0049] The present invention discloses a multi-spectral sensor system for detecting the surface of a high-temperature special steel ingot, comprising: an image acquisition unit, an FPGA module, a GPU module, a PCI E bus, a CPU module, a DDR module and a transmission module; wherein:

[0050] An image acquisition unit collects multi-spectral images of the steel ingot surface and sends them to the FPGA module;

[0051] The FPGA module processes the multispectral image of the steel ingot surface and sends the processed multispectral image to the GPU module;

[0052] The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module;

[0053] The CPU module reads the multispectral image from the DDR module and sends it to the transmission module;

[0054] Transmission module,sends multispectral images to external servers.

[0055] Preferably, the image acquisition unit includes a multispectral optical component and an imaging sensitive unit; the multispectral optical component is used to acquire the multispectral image of the steel ingot surface; the imaging sensitive unit is used to convert the multispectral image of the steel ingot surface into an electrical signal.

[0056] Preferably, the multi-spectral optical component comprises a mounting plate, an R-band lens assembly, a G-band lens assembly, a B-band lens assembly and a far-infrared band lens assembly; wherein the R-band lens assembly, the G-band lens assembly, the B-band lens assembly and the far-infrared band lens assembly are evenly distributed in a square, and the angles between the optical axes of the R-band lens assembly, the G-band lens assembly, the B-band lens assembly and the far-infrared band lens assembly and the normal of the mounting plate are all 15 to 25 degrees.

[0057] Preferably, the R-band lens assembly is provided with a filter with a spectral range of 645 to 655 nm; the G-band lens assembly is provided with a filter with a spectral range of 525 to 535 nm; and the B-band lens assembly is provided with a filter with a spectral range of 445 to 455 nm.

[0058] Preferably, the FPGA module includes three visible light image processing modules, an infrared image processing module and a synchronization module; wherein the synchronization module sends a trigger signal to the three visible light image processing modules and the infrared image processing module; the three visible light image processing modules and the infrared image processing modules perform image acquisition and image processing according to the trigger signal, and send the processed multispectral image to the GPU module.

[0059] Preferably, the visible light image processing module includes a visible light image acquisition module, a bad pixel correction module, a visible light filtering processing module, a grayscale conversion module, a visible light correction module and an image enhancement module; the image is sequentially acquired, bad pixel corrected, filtered, grayscale converted, image corrected and enhanced to generate a visible light image, which is sent to the GPU module.

[0060] Preferably, the infrared image processing module includes: an infrared image acquisition module, a non-uniformity correction module, an infrared filter processing module, a DDE processing module and a temperature calculation module; wherein,

[0061] The non-uniformity correction module performs shutter correction on the data collected by the infrared image acquisition module, and performs KB correction on the shutter-corrected data to obtain a uniform RAW image; the infrared filtering processing module performs filtering on the RAW image and sends it to the DDE processing module and the temperature calculation module; the DDE processing module performs image enhancement on the RAW image after filtering, obtains the enhanced image, and sends it to the GPU module; the temperature calculation module performs temperature calibration and calculation on the RAW image after filtering, obtains temperature data, and sends it to the GPU module.

[0062] Preferably, the DDE processing module includes a high- and low-frequency separation processing module, an adaptive local enhancement module, and an infrared correction module; wherein,

[0063] The high- and low-frequency separation processing module performs Gaussian filtering on the filtered RAW image to obtain a low-frequency image, and subtracts the low-frequency image from the filtered RAW image to obtain a high-frequency image; the adaptive local enhancement module performs adaptive local enhancement on the low-frequency image and the high-frequency image, fuses the enhanced high-frequency image and the low-frequency image and converts them into a grayscale image; the infrared correction module corrects the grayscale image and sends the corrected grayscale image to the GPU module.

[0064] A method for detecting a high-temperature special steel ingot surface using a multi-spectral sensor system comprises:

[0065] The image acquisition unit acquires multi-spectral images of the steel ingot surface and sends them to the FPGA module;

[0066] The FPGA module processes the multispectral image of the steel ingot surface and sends the processed multispectral image to the GPU module;

[0067] The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module;

[0068] The CPU module reads the multispectral image from the DDR module and sends it to the transmission module;

[0069] The transmission module sends the multispectral images to an external server.

[0070] Example

[0071] The present invention discloses a multispectral sensor system for high-temperature special steel ingot surface detection. The multispectral sensor system uniquely adopts multiple image sensor chips, which perform synchronous imaging in different spectral bands, thereby realizing the acquisition of multispectral images of the surface of the special steel ingot. In order to process these multispectral images, the present invention adopts a heterogeneous high-speed computing processing unit based on FPGA, GPU and CPU. This processing unit can realize synchronous acquisition and high-speed parallel computing, thereby meeting the acquisition and processing requirements of high dynamic and high concurrent data volumes. The block diagram of the composition of the multispectral sensor system for high-temperature special steel ingot surface detection is shown in FIG. Figure 1 shown.

[0072] The multi-spectral sensor system designed by the present invention includes three large-array black-and-white image sensor chips (array is 4512x4096) and one thermal imaging image sensor chip (640x480). This design realizes multi-modal multi-spectral imaging, thereby providing more comprehensive and detailed surface information of special steel ingots. In order to further improve the accuracy and stability of detection, the three large-array image sensors in the present invention also use narrow-band multi-spectral imaging technology. This technology can effectively eliminate environmental interference and eliminate the influence of factors such as high humidity and aerosol on the detection results.

[0073] Therefore, the multispectral sensor system for high-temperature special steel ingot surface detection designed by the present invention can not only provide high-quality special steel ingot surface images, but also meet the defect detection requirements of high-speed ingots. The application of this multispectral sensor system will have an important impact on improving the production efficiency and product quality of special steel ingots.

[0074] A multi-spectral sensor system for detecting the surface of a high-temperature special steel ingot of this embodiment is as follows: Figure 1 As shown, the system includes components such as an image acquisition unit, an FPGA module, a GPU module, a PCI E bus, a CPU module, a DDR module and a transmission module, wherein the image acquisition unit acquires a multispectral image of the surface of the steel ingot and sends it to the FPGA module. The FPGA module processes the multispectral image of the surface of the steel ingot and sends the processed multispectral image to the GPU module. The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module. The CPU module reads the multispectral picture from the DDR module and sends it to the transmission module. The transmission module sends the multispectral picture to an external server.

[0075] The image acquisition unit is composed of a multi-spectral optical component and an imaging sensitive unit. The block diagram of the image acquisition unit is as follows: Figure 2 As shown. When the multispectral sensor system is imaging, it is necessary to configure specific multispectral optical components. The multispectral optical components include a mounting plate, an R-band lens assembly, a G-band lens assembly, a B-band lens assembly, and a far-infrared band lens assembly. The mounting plate is used to fix the R-band lens assembly, the G-band lens assembly, the B-band lens assembly, and the far-infrared band lens assembly. The mounting plate arranges the four lens assemblies in a square position and fixes the four lens assemblies at an angle of 15 to 25 degrees to the normal.

[0076] The R-band lens assembly is mainly composed of lens 1, filter 2 and structure. Lens 1 is a high-definition visible light lens. Filter 2 is a narrow-band filter of 645-655nm. The R-band lens assembly completes 645-655nm spectral imaging through lens 1 and filter 2.

[0077] The G-band lens assembly is mainly composed of a lens 6, a filter 5 and a structure. The lens 6 is a high-definition visible light lens. The filter 5 is a narrow-band filter of 525 to 535 nm. The G-band lens assembly completes 525 to 535 nm spectral imaging through the lens 6 and the filter 5.

[0078] The B-band lens assembly is mainly composed of a lens 7, a filter 8 and a structure. The lens 7 is a high-definition visible light lens. The filter 8 is a narrow-band filter of 445 to 455 nm. The B-band lens assembly completes 445 to 455 nm spectral imaging through the lens 7 and the filter 8.

[0079] The far-infrared band lens assembly is mainly composed of an 8-14um far-infrared lens 11, and the far-infrared lens 11 is coated with an 8-14um anti-reflection film to complete 8-14um spectral imaging.

[0080] The design scheme of the filter bandwidth in the multi-spectral optical component is to select a frequency point with relatively high energy according to the spectral energy distribution of the blue spectrum, green spectrum and red spectrum in the visible light band to achieve a narrow band range as small as possible. Through analysis and calculation, the bandwidth of the narrow-band filter of the present invention is 10nm, and the spectrum ranges are 445-455nm, 525-535nm, and 645-655nm respectively, which can effectively eliminate environmental interference and eliminate the influence of factors such as high humidity and fog on the detection results on the spot. The imaging effect is as follows Figure 4 , Figure 5 , Figure 6 , Figure 7 shown.

[0081] The imaging sensitive unit converts the optical signal obtained by the multi-spectral optical component into an electrical signal to achieve image formation. In order to meet the recognition of 100um defects in a 200mm*200mm area, the imaging sensitive unit uses three large arrays of black and white image sensor chips, the array is 4512x4096, and the optical resolution can reach 48um. These sensor chips can image in the spectrum of 445~455nm, 525~535nm, and 645~655nm through the multi-spectral optical component.

[0082] Three large arrays of black and white image sensor chips ( Figure 2 3, 4, 9) have the characteristics of high-speed global exposure and high frame rate. The exposure time can be as low as 1us and the frame rate can be 75fps, which can obtain real-time multispectral images of special steel surfaces.

[0083] At the same time, a thermal imaging sensor chip ( Figure 2 10) with a resolution of 640x480. This sensor chip is used to perform imaging in the 8-14um spectrum to obtain a thermal imaging image of the surface of the special steel ingot. In addition, the imaging sensitive unit also includes a power supply and a driver for the corresponding chip.

[0084] In view of the large amount of multispectral data, the multispectral sensor system designed by the present invention adopts a multi-core heterogeneous solution based on FPGA module, GPU module and CPU module to perform high-speed parallel processing of multispectral data (such as Figure 1 ). Among them, the FPGA module is mainly responsible for synchronous image acquisition and preliminary image processing; the GPU is responsible for image registration and encoding; and the CPU is responsible for decision-making and control.

[0085] like Figure 3 As shown in the figure, the FPGA module collects the real-time image of the image acquisition unit for processing, and transmits it to the GPU module through the PCIE module after the processing is completed. The FPGA module includes three visible light image processing modules, an infrared image processing module and a synchronization module. The synchronization module outputs synchronization signals to the three visible light image processing modules and the infrared image processing module. The three visible light processing modules and the infrared processing module synchronously and parallelly process their corresponding multi-spectral images. After the 4-way multi-spectral image processing is completed, it is transmitted to the GPU module through the PCIE module.

[0086] The synchronization module connects three visible light image processing modules and infrared image processing modules to achieve high-precision synchronous acquisition of images. The synchronization module can generate high-precision synchronization signals based on external trigger signals or internal trigger signals. The synchronization module generates 4 nanosecond precision synchronization pulse signals, which are respectively connected to image acquisition A in visible light image processing module A, image acquisition B in visible light image processing module B, image acquisition C in visible light image processing module C, and image acquisition D in infrared image processing module. The wiring inside the synchronization module is strictly implemented with equal length to ensure the accuracy of the synchronization signal.

[0087] The visible light image processing module A is connected to the R-band multispectral output image (645-655nm spectral image) in the image acquisition unit, and mainly completes the R-band multispectral image acquisition, image noise reduction and image enhancement functions. The visible light image processing module A includes image acquisition A, bad pixel correction A, filtering processing A, grayscale conversion A, image correction A and image enhancement A.

[0088] The visible light image processing module B is connected to the G-band multispectral output image (525-535nm spectral image) in the image acquisition unit, and mainly completes the acquisition of the G-band multispectral image, image noise reduction and image enhancement. The visible light image processing module B includes image acquisition B, bad pixel correction B, filtering processing B, grayscale conversion B, image correction B and image enhancement B.

[0089] The visible light image processing module C is connected to the B-band multispectral output image (445-455nm spectral image) in the image acquisition unit, and mainly completes the acquisition of the B-band multispectral image, image noise reduction and image enhancement. The visible light image processing module C includes image acquisition C, bad pixel correction C, filtering processing C, grayscale conversion C, image correction C and image enhancement C.

[0090] The infrared image processing module is connected to the infrared 8-14um multi-spectral output image in the image acquisition unit, and mainly completes the functions of infrared image acquisition, non-uniformity correction, image noise reduction, temperature calculation and DDE image enhancement. The infrared image processing module includes image acquisition D, non-uniformity correction, filtering processing D, DDE processing and temperature calculation. Among them, DDE processing includes high and low frequency separation processing, adaptive local enhancement and image correction D implementation.

[0091] Image acquisition A realizes image acquisition of 645-655nm spectral imaging chip, and collects image data of the image sensor chip in real time through multiple LVDS data lines. The module realizes synchronous exposure imaging through the synchronous trigger signal of the synchronization module. The module outputs 10-bit RAW format image data of 4512x4096 pixels.

[0092] The bad pixel correction A removes the bad pixels in the image output by the image acquisition A. The bad pixel correction A module replaces the bad pixels previously calibrated by the spectral imaging chip with neighboring pixels, thereby achieving efficient bad pixel correction.

[0093] The filtering process A performs filtering on the RAW format image data output by the bad pixel correction A. Considering the large amount of image data, in order to balance speed and performance, a 3x3 median filtering process is used here to reduce noise and filter out high-frequency noise and dynamic bad pixels.

[0094] The grayscale conversion A performs grayscale conversion on the RAW format image data output after the filtering process A, and converts it into a 256-color grayscale image.

[0095] Image correction A performs gamma correction on the grayscale image output by grayscale conversion A. Image correction A implements fast gamma correction by looking up the gamma values ​​of grayscale values ​​from 0 to 255 in a table, where the gamma correction coefficient is 1.2. Image correction A improves the contrast and brightness of the image.

[0096] Image enhancement A performs Sobel edge enhancement on the image output by image correction A. This step aims to enhance the edge information in the image, especially the defects and cracks on the surface of the high-temperature special steel ingot, thereby significantly improving the visibility of these features and providing a clearer image basis for subsequent defect detection and analysis.

[0097] The image acquisition, bad pixel correction, filtering processing, grayscale conversion, image correction, and image enhancement in the visible light image processing module B and the visible light image processing module C are basically the same as the corresponding sub-module functions in the visible light image processing module A and will not be described here.

[0098] Image acquisition D realizes image acquisition of 8-14um spectral imaging chip. The image acquisition D module generates the configuration clock and configuration signal required by the chip, reads the 14-bit RAW data output by the chip, and outputs 640x480 pixels of 14-bit RAW data for subsequent processing.

[0099] Non-uniformity correction NUC correction is performed on the 14-bit RAW data after image acquisition D. First, shutter correction is performed on the received data, and secondly, KB correction is performed on the shutter-corrected data. After KB correction is completed, the non-uniformity between image pixels is eliminated, and a frame of uniform 14-bit RAW image can be obtained.

[0100] Filtering D performs two filtering operations on the corrected image output by the non-uniformity correction. First, the image is subjected to a 3x3 median filter to filter out high-frequency interference in the image and remove some bad pixels in the image. Then, the image after the median filter is subjected to a bilateral filter to minimize noise and preserve edges.

[0101] The high-low frequency separation process performs a 3x3 Gaussian filter on the filtered image after the filtering process D, and obtains the low-frequency image layer L in the image after filtering. At the same time, the low-frequency image layer L is subtracted from the filtered image after the filtering process D to obtain the high-frequency image layer H of the image. After the high-low frequency separation process, the high-frequency and low-frequency images in the image are separated, which is convenient for the subsequent image enhancement process.

[0102] Adaptive local enhancement performs CLANE enhancement with different parameters on the low-frequency image and high-frequency image after high- and low-frequency separation. After the enhancement is completed, the high- and low-frequency images are fused and converted into grayscale images.

[0103] Image correction D performs gamma enhancement processing on the image after adaptive local enhancement, where the gamma correction coefficient is 2.2 to optimize the contrast and detail performance of the image.

[0104] Temperature calculation performs temperature calibration and calculation on the filtered image to obtain temperature data of 640x480 points. The temperature data can be used for temperature compensation in an external server.

[0105] The above are the detailed functions of the FPGA module.

[0106] The GPU module performs high-speed pixel-based four-eye image registration on the images processed by the FPGA module. After the registration is completed, the four images are respectively encoded in high-speed JPEG parallel encoding. After the encoding is completed, the data is transmitted to the high-speed DDR module through the PCI E bus.

[0107] The CPU module packages the data in the high-speed DDR module through the PCI E bus and transmits it to the transmission module in real time. After receiving the data, the transmission module transmits the data to the external server through the efficient and reliable UDP communication protocol to complete the data transmission of the multispectral sensor.

[0108] Through the above steps, a multispectral sensor for high-temperature special steel ingot surface detection is designed. The multispectral sensor designed by this method can detect surface defects of special steel ingots in real time and accurately under harsh on-site environments.

[0109] The present invention adopts a heterogeneous high-speed computing processing method based on FPGA module, GPU module and CPU module to realize high-speed parallel computing of synchronous acquisition, image processing, image coding and data packaging processing. Among them, the FPGA module is mainly responsible for synchronous image acquisition and image processing; the GPU module is responsible for image registration and image coding; the CPU module is responsible for decision-making and control. The FPGA module, GPU module and CPU module exchange data through a high-speed PC IE data bus to further improve the speed and efficiency of data processing. The efficient data bus can ensure that a large amount of multispectral image data can be quickly transmitted from the FPGA module to the CPU module and the CPU module, thereby realizing the high dynamic image detection capability of the multispectral sensor.

[0110] The design of this embodiment can provide high-quality surface images of high-temperature special steel ingots and meet the defect detection requirements under harsh on-site environments.

[0111] The above embodiments are only explanations of the present invention and cannot be used as limitations of the present invention. Therefore, all implementation methods similar to the concept of the present invention or implementation methods used in other similar structures but with similar concepts to the present invention are within the protection scope of the present invention.

[0112] The above description is only the best specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

[0113] The contents not described in detail in the specification of the present invention belong to the common knowledge of the professionals in this field.

Claims

1. A multi-spectral sensor system for high-temperature special steel ingot surface detection, characterized in that: include: Image acquisition unit, FPGA module, GPU module, PCIE bus, CPU module, DDR module and transmission module; wherein, An image acquisition unit collects multi-spectral images of the steel ingot surface and sends them to the FPGA module; The FPGA module processes the multispectral image of the steel ingot surface and sends the processed multispectral image to the GPU module; The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module; The CPU module reads the multispectral image from the DDR module and sends it to the transmission module; Transmission module,sends multispectral images to external servers.

2. A multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 1, characterized in that: The image acquisition unit comprises a multispectral optical component and an imaging sensitive unit; the multispectral optical component is used to acquire a multispectral image of the surface of the steel ingot; the imaging sensitive unit is used to convert the multispectral image of the surface of the steel ingot into an electrical signal.

3. A multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 2, characterized in that: The multi-spectral optical component comprises a mounting plate, an R-band lens assembly, a G-band lens assembly, a B-band lens assembly and a far-infrared band lens assembly; wherein the R-band lens assembly, the G-band lens assembly, the B-band lens assembly and the far-infrared band lens assembly are evenly distributed in a square, and the angles between the optical axes of the R-band lens assembly, the G-band lens assembly, the B-band lens assembly and the far-infrared band lens assembly and the normal of the mounting plate are all 15 to 25 degrees.

4. A multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 3, characterized in that: The R-band lens assembly is provided with a filter with a spectral range of 645 to 655 nm; the G-band lens assembly is provided with a filter with a spectral range of 525 to 535 nm; and the B-band lens assembly is provided with a filter with a spectral range of 445 to 455 nm.

5. The multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 1 is characterized in that: The FPGA module includes three visible light image processing modules, an infrared image processing module and a synchronization module; wherein the synchronization module sends a trigger signal to the three visible light image processing modules and the infrared image processing module; the three visible light image processing modules and the infrared image processing module perform image acquisition and image processing according to the trigger signal, and send the processed multispectral image to the GPU module.

6. A multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 5, characterized in that: The visible light image processing module includes a visible light image acquisition module, a bad pixel correction module, a visible light filtering processing module, a grayscale conversion module, a visible light correction module and an image enhancement module; the image is sequentially acquired, bad pixel corrected, filtered, grayscale converted, image corrected and enhanced to generate a visible light image, which is sent to the GPU module.

7. The multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 5, characterized in that: The infrared image processing module includes: an infrared image acquisition module, a non-uniformity correction module, an infrared filter processing module, a DDE processing module and a temperature calculation module; wherein, The non-uniformity correction module performs shutter correction on the data collected by the infrared image acquisition module, and performs KB correction on the shutter-corrected data to obtain a uniform RAW image; the infrared filtering processing module performs filtering on the RAW image and sends it to the DDE processing module and the temperature calculation module; the DDE processing module performs image enhancement on the RAW image after filtering, obtains an enhanced image, and sends it to the GPU module; the temperature calculation module performs temperature calibration and calculation on the RAW image after filtering, obtains temperature data, and sends it to the GPU module.

8. A multi-spectral sensor system for detecting the surface of high-temperature special steel ingot according to claim 7, characterized in that: The DDE processing module includes a high- and low-frequency separation processing module, an adaptive local enhancement module, and an infrared correction module; wherein, The high- and low-frequency separation processing module performs Gaussian filtering on the filtered RAW image to obtain a low-frequency image, and subtracts the low-frequency image from the filtered RAW image to obtain a high-frequency image; the adaptive local enhancement module performs adaptive local enhancement on the low-frequency image and the high-frequency image, fuses the enhanced high-frequency image and the low-frequency image and converts them into a grayscale image; the infrared correction module corrects the grayscale image and sends the corrected grayscale image to the GPU module.

9. A method for detecting a high-temperature special steel ingot surface using a multi-spectral sensor system as claimed in any one of claims 1 to 8, characterized in that: include: The image acquisition unit acquires multi-spectral images of the steel ingot surface and sends them to the FPGA module; The FPGA module processes the multispectral image of the steel ingot surface and sends the processed multispectral image to the GPU module; The GPU module performs image registration and encoding on the processed multispectral image to form a multispectral picture and sends it to the DDR module; The CPU module reads the multispectral image from the DDR module and sends it to the transmission module; The transmission module sends the multispectral images to an external server.

Images