Color sorter system based on double hyperspectral cameras

Abstract

The invention relates to the technical field of color sorters, and discloses a color sorter system based on double hyperspectral cameras, which comprises a hopper, a vibration feeder, a crawler belt conveyor belt, a hyperspectral camera detection module, an air injection system, a finished product bin and a waste bin, wherein the hopper is used for inputting materials; the vibration feeder is used for uniformly conveying materials to the crawler belt conveyor belt; the crawler belt conveyor belt is used for receiving materials conveyed by the vibration feeder and continuously conveying the materials forwards; and the hyperspectral camera detection module is used for carrying out image acquisition on the material under a uniform illumination condition. Compared with a color selector based on an RGB camera, the color selector provided by the invention has more accurate spectral characteristic representation, can obtain more and more comprehensive spectral information data and has more application scenes and development spaces, and compared with a color selector based on a near-infrared camera, the color selector provided by the invention has fewer cameras and optical filters, the structure is simpler, and the wave band selection is more convenient.

Description

Technical Field

[0001] The present invention relates to the technical field of color sorters, and in particular to a color sorter system based on dual hyperspectral cameras. Background Art

[0002] Color sorter is a device used in food, medicine, metal products, minerals, chemicals and other industries to automatically identify and sort items of different colors or shapes. It uses advanced image recognition technology and machine vision technology to quickly and accurately identify items of different colors, shapes and sizes and realize automatic sorting.

[0003] At present, the main components of the color sorter include:

[0004] 1. Feeding mechanism and discharging mechanism: The feeding mechanism is used to hold the materials before sorting and make them in a uniform distribution state, including the feeding hopper and the vibrator; the discharging mechanism is used to hold the materials after sorting, including the finished product hopper and the waste hopper;

[0005] 2. Conveying mechanism: used to convey materials to the photoelectric detection mechanism and the actuator, and at the same time shape the materials into a single layer and evenly distributed state, including conveyor belts and drive motors, etc.;

[0006] 3. Actuator: According to the sorting signal output by the image processing mechanism, the actuator is driven to perform the material sorting operation, including high-frequency solenoid valve, blowing nozzle and air pipe, etc.;

[0007] 4. Photoelectric detection mechanism: used to collect image information of materials and transmit the image information to the image processing mechanism, including light source, camera and lens;

[0008] 5. Image processing mechanism: processes and analyzes the collected image information, identifies materials of different colors, shapes and sizes, and outputs sorting signals to achieve sorting, including image processing software and human-computer interface.

[0009] The existing color sorting machines are mainly based on RGB cameras for imaging. RGB cameras can identify the three bands of R, G, and B, and can obtain the reflected light of the target material at these three bands. Then, through the image fusion of the three bands, the color information of the target material is obtained and the corresponding two-dimensional image is synthesized. However, the three bands of R, G, and B recognized by this RGB camera have crosstalk, such as Figure 1 As shown, channel crosstalk will cause overlap between the spectral responses of the red, green, and blue channels. The more severe the crosstalk between channels is, the more the corresponding spectral lines overlap, and the greater the uncertainty of the obtained spectral data is.

[0010] Moreover, RGB cameras only cover three visible light bands, R, G, and B, corresponding to the red, green, and blue channels respectively. Their recognition capabilities are limited. Compared with hyperspectral cameras, they cover fewer bands and cannot see the spectral information of bands other than the visible light band. They have certain limitations and are not suitable for special materials such as oil content detection in nut kernels, which is not conducive to further analysis of spectral data.

[0011] The hyperspectral camera, also known as the hyperspectral imaging system, is mainly composed of a front lens, a slit, a spectral spectrometer system, and a detector. According to the different spectral spectrometer systems, push-broom hyperspectral cameras are classified into dispersion type, interference type, filter type, and computational tomography type. Among them, the dispersion type imaging spectrometer has good spectral linearity, is widely used, and has mature technology.

[0012] Compared with the traditional color sorter based on RGB camera, the visible-near infrared hyperspectral camera can directly obtain the R, G, and B band information respectively without the influence of spectral line overlap. More generally, the hyperspectral camera can obtain the full-band spectral information at any pixel of the two-dimensional image without the influence of spectral line overlap.

[0013] Compared with the existing color sorters based on near-infrared cameras, if you want to select images of a specific band for analysis, the general practice of color sorters is to use corresponding narrow-band filters to filter out the near-infrared spectral information of the corresponding wavelength. If you want to select three bands at the same time, you need to use three near-infrared cameras plus corresponding filters. This approach is not only costly, but also the size of the color sorter is relatively large, and the structure is not very compact. The selection of a specific wavelength can only be achieved by increasing the number of cameras and filters. However, the near-infrared hyperspectral camera can achieve the selection of several bands within the wavelength range at the same time, which is relatively more convenient.

[0014] The combination of the visible-near infrared hyperspectral camera and the near infrared hyperspectral camera of the present invention can realize the sorting of the shapes and colors of objects, as well as the sorting of the materials and qualities of objects. Summary of the invention

[0015] In order to solve the technical problems raised in the background technology, the present invention proposes a color sorter system based on dual hyperspectral cameras, including a hopper, a vibrating feeder, a crawler conveyor, a hyperspectral camera detection module, an air jet system, a finished product silo and a waste silo; wherein:

[0016] The hopper is used to input materials;

[0017] The vibrating feeder is used to evenly feed the material to the crawler conveyor belt;

[0018] The crawler conveyor is used to receive the material delivered by the vibrating feeder and continue to deliver it forward;

[0019] The hyperspectral camera detection module collects images of materials under uniform lighting conditions, and can synchronously collect spatial information of the scanning dimension z through the movement of the crawler conveyor belt, and performs subsequent algorithm processing on the collected images through the corresponding image processing system;

[0020] The jet system is used to blow unqualified materials located by the hyperspectral camera detection module toward the waste bin;

[0021] The finished product silo is used to receive qualified materials;

[0022] The waste bin is used to receive unqualified materials;

[0023] The image processing system includes four functional modules: wavelength calibration module, spectrum bending correction module, black and white correction module and image reconstruction module;

[0024] The wavelength calibration module is used to perform calibration based on a mercury-argon lamp. The mercury-argon lamp illuminates the slit of the spectrum splitting system to obtain the pixel-wavelength position calibration relationship between the slit spectrum line on the detector and the corresponding mercury-argon lamp wavelength.

[0025] The spectral bending correction module is used to algorithmically correct the spectral bending and color distortion generated by the optical system itself through non-uniform linear interpolation and normal mapping methods;

[0026] The black and white correction module is used to eliminate the effects of uneven illumination, uneven reflectivity of the calibration white board, dark current noise, etc. during image acquisition, and pre-process the raw data of the hyperspectral image;

[0027] The image reconstruction module is used for image display. It splices together a hyperspectral three-dimensional image synthesized by spectral dimension x, spatial dimension y and scanning dimension z. Based on wavelength calibration, several bands are selected to synthesize a pseudo-color image to obtain the spectral characteristic curve of each pixel in the pseudo-color image.

[0028] Preferably, the hyperspectral camera detection module includes a visible-near infrared hyperspectral camera, a near infrared hyperspectral camera, and a white light LED light source and a halogen light source respectively arranged adjacent to the two;

[0029] The visible-near infrared hyperspectral camera operates in the wavelength range of 400-1000nm;

[0030] The operating band of the near-infrared hyperspectral camera is 900-1700nm;

[0031] Preferably, the optical path structure of the visible-near infrared hyperspectral camera includes a slit 1 with a length of 10 mm, a collimating lens 1, a prism 1, a transmission diffraction grating 1, a focusing lens 1, and a sensor receiving surface 1, which are arranged in sequence from left to right; the collimating lens 1 and the focusing lens 1 are both aspheric doublet lenses with the same focal length of 50 mm;

[0032] The optical path structure of the visible-near infrared hyperspectral camera also includes a front telephoto lens for imaging arranged in front of the slit, and the focal length of the front telephoto lens is optional at 16mm / 25mm / 35mm, and the working distance of the front telephoto lens is the distance from the front end of the lens to the material sorted by the color sorter. The corresponding front telephoto lens can be configured according to different working distances, such as a front lens with a focal length of 16mm at a working distance of 300mm.

[0033] Preferably, the optical path structure of the near-infrared hyperspectral camera includes two slits, two collimating lenses, two prisms, two transmission diffraction gratings, two focusing lenses, and two sensor receiving surfaces, which are arranged in sequence from left to right; and the optical path structure of the near-infrared hyperspectral camera is different from that of the visible-near-infrared hyperspectral camera in that the focusing lens two includes: a best shape lens with a focal length of 200 mm, a positive meniscus lens with a focal length of 200 mm, and an aspherical single lens with a focal length of 50 mm.

[0034] The optical path structure of the near-infrared hyperspectral camera also includes a front telephoto lens for imaging set in front of the slit, and the focal length of the front telephoto lens is optional at 16mm / 25mm / 35mm. The working distance of the front telephoto lens is the distance from the front end of the lens to the material sorted by the color sorter. The corresponding front telephoto lens can be configured according to different working distances, such as a front lens with a focal length of 16mm at a working distance of 300mm.

[0035] In order to reduce the aberration caused by the focusing lens group, the focusing lens group consists of a 200mm best shape lens, a 200mm positive meniscus lens and a 50mm aspherical single lens. This lens group disperses the focal length by combining multiple single lenses, so that the light transition is smoother to reduce spherical aberration. Furthermore, by changing the curvature and surface shape of the spherical lens, an aspherical lens is used to reduce other aberrations such as spherical aberration. Compared with the visible-near-infrared hyperspectral camera, the focusing lens group structure of this near-infrared hyperspectral camera adds two long-focal-length single lenses on the basis of the 50mm aspherical lens to change the magnification of the near-infrared light spectrum splitting system, so that the spectrum magnification is about 0.7.

[0036] Compared with the traditional color sorter based on RGB camera and near-infrared camera, the color sorter based on dual hyperspectral camera can directly obtain the spectral information at any band of 400-1700nm. More generally, the hyperspectral camera can obtain the full-band spectral information of 400-1700nm at any pixel of the two-dimensional image without the influence of spectral line overlap. Ordinary RGB cameras only cover three R, G, and B visible light bands, and have limited recognition capabilities. Hyperspectral cameras can provide richer and more detailed band information, cover wider spectral bands, and have more application scenarios.

[0037] Compared with the existing color sorters based on near-infrared cameras, if you want to select images of a specific band for analysis, the general practice of color sorters is to use corresponding narrow-band filters to filter out the near-infrared spectral information of the corresponding wavelength. If you want to select three bands at the same time, you need to use three near-infrared cameras plus corresponding filters. This approach is not only costly, but also the size of the color sorter is relatively large, and the structure is not very compact. The selection of a specific wavelength can only be achieved by increasing the number of cameras and filters. However, the near-infrared hyperspectral camera can achieve the selection of several bands within the wavelength range at the same time, which is relatively more convenient.

[0038] The dual hyperspectral camera of the present invention has a wavelength range of 400-1000nm and 900-1700nm, which can be used to quantitatively analyze the 400-1700nm spectrum. The color sorter based on the dual hyperspectral camera can be used not only to detect the shape and color of the material, but also to detect the material and quality characteristics of the material, such as chemical properties and nutritional value, sugar, fat and protein content, etc.

[0039] It can obtain more comprehensive spectral information data than color sorters based on RGB cameras and near-infrared cameras. Ordinary RGB cameras and near-infrared cameras can only perform simple quantitative analysis, while hyperspectral cameras can perform more detailed quantitative analysis and can be used in a wider range of applications. In addition, the selection of bands by near-infrared cameras requires multiple cameras and corresponding filters, which are relatively complex in structure and costly, and it is cumbersome to replace filters each time.

[0040] The hyperspectral cube obtained by the hyperspectral camera is a hyperspectral three-dimensional image synthesized by spectral dimension x, spatial dimension y and scanning dimension z, which is the three-dimensional information of the material to be sorted. Among them, the spatial dimension x represents the one-dimensional position spatial information in the slit direction, the spectral dimension y represents the full-band spectral information within the measurement range, and the scanning dimension z represents the one-dimensional position spatial information in the scanning direction; the two-dimensional image is the two-dimensional position spatial information of the material to be sorted composed of spatial dimension x and scanning dimension z, which is similar to the two-dimensional image collected by the RGB camera or the near-infrared camera, and is extracted from the full spectrum The corresponding band image is fused; the spectral characteristic curve of any pixel is simply to extract a certain pixel point from the two-dimensional spatial position information of the hyperspectral cube, that is, the two-dimensional image, and draw the spectral characteristic curve of the pixel point, that is, the wavelength-spectral reflectance information or wavelength-DN value information of the pixel point. Among them, the wavelength range is the measurable band range of the dual hyperspectral camera of the present invention, and the full band is 400-1700nm; the spectral reflectance is the spectral information obtained by the corresponding image processing algorithm combined with the black and white plate calibration; the DN value is the grayscale value spectral information at the corresponding wavelength.

[0041] Hyperspectral cameras are currently developing towards miniaturization and compactness, not only in terms of performance, but also in terms of size. With the continuous improvement of hyperspectral imaging technology, the present invention can be replaced by short-wave infrared or even medium-wave infrared hyperspectral cameras. By integrating the power supply and cooling parts of the hyperspectral camera into the color sorter, the size of the hyperspectral camera can be further reduced, and more, richer and more accurate spectral information can be obtained than ordinary RGB cameras.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention has a more accurate spectral characteristic representation than a color sorter based on an RGB camera; it can obtain more and more comprehensive spectral information data than a color sorter based on an RGB camera; it has more application scenarios and development space than a color sorter based on an RGB camera; it has fewer cameras and filters than a color sorter based on a near-infrared camera, has a simpler structure, and is more convenient for band selection.

[0044] The present invention can simultaneously realize the shape and color sorting of materials as well as the material and quality sorting of materials. The shape and color sorting of materials is mainly carried out by a visible-near infrared hyperspectral camera. The visible-near infrared hyperspectral camera (400-1000nm) can obtain the human visual color of the material under visible light by selecting the corresponding RGB band. The hyperspectral camera can directly take pictures to obtain the appearance characteristics such as the shape and surface texture of the material, so that the shape and color sorting of the material can be realized by the visible-near infrared camera. Unlike ordinary RGB cameras, the visible-near infrared hyperspectral camera can not only realize the shape and color sorting of materials, but also detect the quality characteristics of materials in the near-infrared band, such as the difference in chlorophyll and water content.

[0045] Taking fluorite sorting as an example, by turning on the visible-near infrared hyperspectral camera and the corresponding light source, the hyperspectral camera takes a picture to obtain the corresponding spectral characteristic curve, and selects the best characteristic wavelengths (902.16nm, 571.62nm and 485.14nm) for fluorite detection from the spectral characteristic curve. The components of the three selected characteristic wavelengths are analyzed, and finally the classification of normal fluorite and impurities is obtained through algorithm processing. The clean selection rate of normal fluorite is about 98%.

[0046] Near-infrared hyperspectral cameras (900-1700nm) are mainly responsible for the material and quality sorting of materials. By identifying the near-infrared characteristic peak bands of different materials, different quality characteristics of different materials can be identified, such as the internal organizational information of the materials, nutritional components such as protein and sugar; or different material characteristics of different materials, such as the classification of plastics PE / ABS / PVC / PS / PA / PP, etc.

[0047] Taking pork sorting as an example, by turning on the near-infrared hyperspectral camera and the corresponding light source, the hyperspectral camera takes a picture to obtain the corresponding spectral characteristic curve. From the spectral characteristic curve, we can see that the characteristic wavelength of protein appears around 1200nm, and the characteristic wavelength of fat appears around 1400nm. Based on the visual analysis of these two characteristic bands, the fat and protein content of different pork can be distinguished.

[0048] The present invention is compact in size, simple in structure, and easy to adjust the optical path, suitable for low-cost applications. In addition, the grating in the near-infrared hyperspectral camera does not need to use a holographic grating, because the number of optical elements used is small, and the system directly uses a near-infrared ruled transmission grating with low diffraction efficiency to achieve light splitting to meet the transmittance requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of spectral information obtained by an RGB camera in the prior art;

[0050] Figure 2This is a schematic diagram of the overall structure of the color sorter system based on the hyperspectral camera proposed in the present invention;

[0051] Figure 3a This is a diagram of the optical path structure of the visible-near infrared hyperspectral camera proposed in the present invention;

[0052] Figure 3b This is a diagram of the optical path structure of the near-infrared hyperspectral camera proposed in the present invention;

[0053] Figure 4 This is a schematic diagram of spectral information acquired by the hyperspectral camera proposed in the present invention;

[0054] Figure 5 A three-dimensional illustration of the hyperspectral camera proposed in this invention.

[0055] Description of main symbols:

[0056] Material 1; feeding hopper 2; vibrating feeder 3; crawler conveyor 4; visible-near infrared hyperspectral camera 5; near infrared hyperspectral camera 6; white light LED light source 7; halogen light source 8; jet system 9; finished product silo 10; waste silo 11;

[0057] Slit 1 51; collimating lens 1 52; prism 1 53; diffraction grating 1 54; focusing lens 1 55; sensor receiving surface 1 56; slit 2 61; collimating lens 2 62; prism 2 63; diffraction grating 2 64; focusing lens 2 65; sensor receiving surface 2 66; optimal shape lens 651, positive meniscus lens 652, aspherical single lens 653. DETAILED DESCRIPTION

[0058] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0059] Example 1: Reference Figure 1-Figure 5 , the present invention includes a hopper, a vibrating feeder, a crawler conveyor, a hyperspectral camera detection module, an air jet system, a finished product silo and a waste silo; wherein:

[0060] The hopper is used to input materials;

[0061] The vibrating feeder is used to evenly feed the material to the crawler conveyor belt;

[0062] The crawler conveyor is used to receive the material delivered by the vibrating feeder and continue to deliver it forward;

[0063] The hyperspectral camera detection module collects images of materials under uniform lighting conditions, and can synchronously collect spatial information of the scanning dimension z through the movement of the crawler conveyor belt, and performs subsequent algorithm processing on the collected images through the corresponding image processing system;

[0064] The jet system is used to blow unqualified materials located by the hyperspectral camera detection module toward the waste bin;

[0065] The finished product silo is used to receive qualified materials.

[0066] The image processing system includes four functional modules: wavelength calibration module, spectrum bending correction module, black and white correction module and image reconstruction module;

[0067] The wavelength calibration module is used to perform calibration based on a mercury-argon lamp. The mercury-argon lamp illuminates the slit of the spectrum splitting system to obtain the pixel-wavelength position calibration relationship between the slit spectrum line on the detector and the corresponding mercury-argon lamp wavelength.

[0068] The spectral bending correction module is used to algorithmically correct the spectral bending and color distortion generated by the optical system itself through non-uniform linear interpolation and normal mapping methods;

[0069] The black and white correction module is used to eliminate the effects of uneven illumination, uneven reflectivity of the calibration white board, dark current noise, etc. during image acquisition, and pre-process the raw data of the hyperspectral image;

[0070] The image reconstruction module is used for image display. It splices together a hyperspectral three-dimensional image synthesized by spectral dimension x, spatial dimension y and scanning dimension z. Based on wavelength calibration, several bands are selected to synthesize a pseudo-color image to obtain the spectral characteristic curve of each pixel in the pseudo-color image.

[0071] Preferably, the hyperspectral camera detection module includes a visible-near infrared hyperspectral camera 5 and a near infrared hyperspectral camera 6 and a white light LED light source 7 and a halogen lamp light source 8 respectively arranged adjacent to the two.

[0072] The visible-near infrared hyperspectral camera 5 operates in the wavelength range of 400-1000nm;

[0073] The optical path structure of the visible-near infrared hyperspectral camera 5 includes a slit 51 with a length of 10 mm, a collimating lens 52, a prism 53, a transmission diffraction grating 54, a focusing lens 55, and a sensor receiving surface 56, which are arranged from left to right in sequence; the collimating lens 52 and the focusing lens 55 are both aspherical double-cemented lenses with the same focal length of 50 mm.

[0074] The optical path structure of the visible-near infrared hyperspectral camera 5 also includes a front telephoto lens for imaging arranged in front of the slit, and the focal length of the front telephoto lens is optional at 16mm / 25mm / 35mm. The working distance of the front telephoto lens is the distance from the front end of the lens to the material sorted by the color sorter. The corresponding front telephoto lens can be configured according to different working distances, such as a front lens with a focal length of 16mm at a working distance of 300mm.

[0075] like Figure 3a As shown, the specific working principle is as follows: the target to be measured is imaged on the slit 51 through the front telephoto lens, and the image of the line area passing through the slit 51 becomes parallel light through the collimating lens 52. The parallel light is incident on the first beam splitting element prism 53 and is split once by the prism 53 and changes the light emission angle. After the first split, the light is transmitted through the second beam splitting element diffraction grating 54, and is split twice by the diffraction grating 54. Finally, the light of different wavelengths is focused on the different spectral dimensions of the sensor receiving surface 56 through the focusing lens 55, and the material characteristics are restored according to the corresponding image processing algorithm. Here, the front telephoto lens adopts the common object space telecentric lens on the market. The focal length of the front telephoto lens is optional at 16mm / 25mm / 35mm. The working distance of the front telephoto lens is the distance from the front end of the lens to the material sorted by the color sorter. The corresponding front telephoto lens can be configured according to different working distances, such as configuring a front lens with a focal length of 16mm at a working distance of 300mm. Correspondingly, the double-cemented lens 52 and the double-cemented lens 55 are both aspheric double-cemented lenses with the same focal length of 50 mm. Correspondingly, a corresponding bandpass filter is installed on the front lens, and the bandpass range is 400-1000 nm.

[0076] Embodiment 2:

[0077] On the basis of the visible-near infrared hyperspectral spectroscopic system, the present invention further proposes a hyperspectral spectroscopic system suitable for near infrared, with a wavelength range of 900-1700nm.

[0078] The optical design method of near-infrared is similar to that of visible light. The difference is that since the spectral range of the near-infrared hyperspectral spectrometer is wider, the focusing lens group needs to be processed by optical focal length splitting and aspheric surface conversion to reduce aberrations.

[0079] Reference Figure 3bThe near-infrared hyperspectral spectroscopic system is composed of a slit 2 61, a doublet lens 2 62, an equilateral prism 2 63, a diffraction grating 2 64, a focusing lens 2 65, a sensor receiving surface 2 66 and a filter. Among them, the equilateral prism 2 63 and the diffraction grating 2 64 constitute a prism assembly. Correspondingly, the focusing lens 2 65 includes: a 200mm best shape lens 651, a 200mm positive meniscus lens 652 and a 50mm aspherical single lens 653. Here, the optical devices are all coated, and the film layer transmission range is in the near-infrared range of 900-1700nm.

[0080] The target to be measured is imaged on the slit 2 61 through the front telephoto lens. The image of the line area passing through the slit 2 61 becomes parallel light through the collimating lens 2 62. The parallel light is incident on the first beam splitter element prism 2 63 and is split once by the prism 2 63 and changes the light emission angle. After the first split, the light is transmitted through the second beam splitter element diffraction grating 2 64, and is split twice by the diffraction grating 2 64. Finally, the light of different wavelengths is focused on different spectral dimensions of the sensor receiving surface 2 66 through the focusing lens group 2 65, and the material characteristics are restored according to the corresponding image processing algorithm. Here, the front telephoto lens adopts the common object telecentric lens on the market, and the working distance is the distance from the front end of the lens to the color sorter to sort the material. Correspondingly, the corresponding bandpass filter is installed on the front lens, and the bandpass range is 900-1700nm. Accordingly, in order to reduce the aberration generated by the focusing lens group 2 65, the focusing lens group 2 65 is composed of a 200 mm best shape lens 651, a 200 mm positive meniscus lens 652, and a 50 mm aspherical single lens 653. This combination disperses the focal length by using multiple single lenses, so that the light transition is smoother to reduce spherical aberration, and by changing the curvature and surface shape of the spherical lens, using the aspherical lens to reduce spherical aberration and other aberrations, and by adding two long focal length single lenses on the basis of the 50 mm aspherical lens to change the magnification of the spectrum splitting system, the spectrum magnification is about 0.7.

[0081] The optical system design method proposed in the present invention is as follows: first determine the technical indicators of the system, calculate the initial parameters according to the technical indicators, including the size of the slit width and length, the focal length of the collimating lens and the focusing lens, the number of lines of the diffraction grating, the angle of the equilateral prism, the size of the image plane, etc., and select the slit, collimating lens, prism, diffraction grating, focusing lens and detector according to the calculated parameters. Finally, perform ZEMAX simulation based on the selected commercially available devices, use the interval, tilt angle, object distance and image distance as variables for optimization, and after determining the position of the optical components, draw and process the corresponding optical and mechanical components to build the experimental platform and verify the results.

[0082] The schematic diagram of the color sorter system based on dual hyperspectral cameras described in the present invention is as follows. Taking a common crawler color sorter as an example, material 1 is poured into the hopper 2, and the vibrating feeder 3 evenly delivers the material 1 to the trough. The trough enters the optical module along the movement direction shown in the figure under the transmission of the crawler conveyor 4, and is evenly illuminated by the white light LED light source 7 and the halogen light source 8. At the same time, the visible-near infrared hyperspectral camera 5 and the near infrared hyperspectral camera 6 collect images of the material 1 under uniform lighting conditions. The spatial information of the scanning dimension z can be synchronously collected by the movement of the crawler conveyor 4, and the collected images are subsequently processed by the corresponding image processing system. The unqualified materials located by the algorithm will be blown to the waste bin 11 by the high-pressure gas instantly ejected by the jet system 9, and when the qualified materials located pass in front of the jet system, no high-pressure gas is blown out, and they will automatically fall to the finished product bin 10 to complete the material sorting. Here, the halogen lamp has a spectrum range of 500-2000nm, which is used as the infrared light source of the near-infrared hyperspectral camera; the white light LED has a spectrum range of 400-750nm, and together with the halogen lamp, it is used as the visible-near-infrared light source of the visible-near-infrared hyperspectral camera. The combination of the two light sources can meet the response requirements of the visible-near-infrared hyperspectral camera and the near-infrared hyperspectral camera.

[0083] In terms of application, the color sorter based on dual hyperspectral cameras described in the present invention combines a visible-near infrared hyperspectral camera and a near infrared hyperspectral camera. The visible-near infrared hyperspectral camera 5 is mainly used for identifying the shape and color of the material, and the near infrared hyperspectral camera 6 is mainly used for detecting other quality characteristics and material characteristics of the material. When the color and quality of the same material need to be sorted, the two hyperspectral cameras need to be used in combination. When only the color and shape of the material need to be sorted, the visible-near infrared hyperspectral camera 5 is turned on and the near infrared hyperspectral camera 6 is turned off. When only the oil content or other quality characteristics of the material need to be sorted, the visible-near infrared hyperspectral camera 5 is turned off and the near infrared hyperspectral camera 6 is turned on.

[0084] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A color sorter system based on dual hyperspectral cameras, characterized in that: It includes a hopper, a vibrating feeder, a crawler conveyor, a hyperspectral camera detection module, an air jet system, a finished product silo, and a waste silo; among which: The hopper is used to input materials; The vibrating feeder is used to evenly feed the material to the crawler conveyor belt; The crawler conveyor is used to receive the material delivered by the vibrating feeder and continue to deliver it forward; The hyperspectral camera detection module collects images of materials under uniform lighting conditions, and can synchronously collect spatial information of the scanning dimension z through the movement of the crawler conveyor belt, and performs subsequent algorithm processing on the collected images through the corresponding image processing system; The jet system is used to blow unqualified materials located by the hyperspectral camera detection module toward the waste bin; The finished product silo is used to receive qualified materials.

2. The color sorter system according to claim 1, characterized in that: The image processing system includes four functional modules: wavelength calibration module, spectrum bending correction module, black and white correction module and image reconstruction module; The wavelength calibration module is used to perform calibration based on a mercury-argon lamp. The mercury-argon lamp illuminates the slit of the spectrum splitting system to obtain the pixel-wavelength position calibration relationship between the slit spectrum line on the detector and the corresponding mercury-argon lamp wavelength. The spectral bending correction module is used to algorithmically correct the spectral bending and color distortion generated by the optical system itself through non-uniform linear interpolation and normal mapping methods; The black and white correction module is used to eliminate the effects of uneven illumination, uneven reflectivity of the calibration white board, dark current noise, etc. during image acquisition, and pre-process the raw data of the hyperspectral image; The image reconstruction module is used for image display. It splices together a hyperspectral three-dimensional image synthesized by spectral dimension x, spatial dimension y and scanning dimension z. Based on wavelength calibration, several bands are selected to synthesize a pseudo-color image to obtain the spectral characteristic curve of each pixel in the pseudo-color image.

3. The color sorter system according to claim 1, characterized in that: The hyperspectral camera detection module includes a visible-near infrared hyperspectral camera, a near infrared hyperspectral camera, and a white light LED light source and a halogen light source respectively arranged adjacent to the two; The visible-near infrared hyperspectral camera operates in the wavelength range of 400-1000nm; The operating band of the near-infrared hyperspectral camera is 900-1700nm.

4. The color sorter system according to claim 1, characterized in that: The optical path structure of the visible-near infrared hyperspectral camera includes a slit 1 with a length of 10 mm, a collimating lens 1, a prism 1, a transmission diffraction grating 1, a focusing lens 1, and a sensor receiving surface 1, which are arranged in sequence from left to right; the collimating lens 1 and the focusing lens 1 are both aspheric doublet lenses with the same focal length of 50 mm; The optical path structure of the visible-near infrared hyperspectral camera also includes a front telephoto lens for imaging arranged in front of slit 1, and the focal length of the front telephoto lens is optional at 16mm / 25mm / 35mm. The working distance of the front telephoto lens is the distance from the front end of the lens to the material sorted by the color sorter, and the corresponding front telephoto lens can be configured according to different working distances.

5. The color sorter system according to claim 1, characterized in that: The optical path structure of the near-infrared hyperspectral camera includes two slits, two collimating lenses, two prisms, two transmission diffraction gratings, two focusing lenses, and two sensor receiving surfaces, which are arranged from left to right in sequence; and the difference between the optical path structure of the near-infrared hyperspectral camera and the optical path structure of the visible-near-infrared hyperspectral camera is that the focusing lens two includes: a best shape lens with a focal length of 200 mm, a positive meniscus lens with a focal length of 200 mm, and an aspherical single lens with a focal length of 50 mm. The optical path structure of the near-infrared hyperspectral camera also includes a front telephoto lens for imaging, which is set in front of the slit. The focal length of the front telephoto lens is optional, 16mm / 25mm / 35mm. The working distance of the front telephoto lens is the distance from the front end of the lens to the material to be sorted by the color sorter. The corresponding front telephoto lens can be configured according to different working distances.

Images