A household garbage identification system

Abstract

The present invention provides a domestic waste identification system, comprising an industrial computer, a waste conveying device, and a color camera, a 3D camera, and a hyperspectral camera arranged in sequence along the conveying direction of the waste conveying device, a linear light source cooperating with the color camera is arranged below the side of the color camera, a laser transmitter cooperating with the 3D camera is arranged next to the 3D camera, a dome light source cooperating with the hyperspectral camera is arranged directly below the hyperspectral camera, and the waste conveying device, the color camera, the 3D camera, and the hyperspectral camera are respectively connected to the industrial computer for communication. The present invention collects multi-dimensional feature information of domestic waste by combining a 3D camera, a color camera, and a hyperspectral camera, and can significantly improve the recognition accuracy of domestic waste through multi-feature information fusion and hierarchical identification, so that machine intelligent sorting can be applied to domestic waste sorting, which helps to improve sorting efficiency and improve sorting conditions.

Description

Technical Field

[0001] The invention relates to a garbage disposal system, in particular to a domestic garbage identification system. Background Art

[0002] With the rapid growth of social economy, people's consumption level is also constantly improving. However, with people's consumption comes hundreds of millions of domestic waste. These domestic wastes are difficult to recycle and can only be landfilled or incinerated, which not only pollutes the environment but also occupies more and more land resources.

[0003] With the introduction of national policies and the improvement of people's environmental awareness, how to better recycle domestic waste and turn waste into treasure has become a hot topic for the government and enterprises. The most important part of recycling domestic waste is sorting, and the current domestic waste sorting method is mainly manual sorting, which has low sorting efficiency and poor sorting conditions. In the face of hundreds of millions of tons of domestic waste, machine intelligent sorting will inevitably replace manual sorting and become a future trend. The first problem that machine intelligent sorting needs to solve is how to identify domestic waste. Although the country has effectively reduced the difficulty of sorting by advocating garbage classification, there are still many types of garbage in each category. Existing machine intelligent sorting equipment can usually only identify specific types of objects, which is difficult to apply to domestic waste sorting.

[0004] In view of this, the applicant conducted an in-depth study on the above-mentioned issues, which led to the present case. Summary of the invention

[0005] The object of the present invention is to provide a household waste identification system which helps to improve sorting efficiency and sorting conditions.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A domestic waste identification system comprises an industrial computer, a waste conveying device, and a color camera, a 3D camera and a hyperspectral camera located above the waste conveying device and arranged in sequence along the conveying direction of the waste conveying device. The waste conveying device, the color camera, the 3D camera and the hyperspectral camera are respectively communicatively connected with the industrial computer.

[0008] As an improvement of the present invention, a linear light source cooperating with the color camera is arranged at the lower side of the color camera, a laser emitter cooperating with the 3D camera is arranged next to the 3D camera, and a dome light source cooperating with the hyperspectral camera is arranged directly below the hyperspectral camera.

[0009] As an improvement of the present invention, a halogen lamp is installed on the dome light source.

[0010] As an improvement of the present invention, the garbage conveying device is provided with a dispersing device for spreading the garbage on the garbage conveying device, and the dispersing device is located on a side of the color camera away from the 3D camera.

[0011] As an improvement of the present invention, the industrial computer has an image algorithm processing module, which screens out dark materials in domestic waste through the image information provided by the color camera and the 3D camera, and obtains the location and category information of non-dark materials through the image information provided by the hyperspectral camera and the 3D camera.

[0012] As an improvement of the present invention, after receiving the image information sent by the color camera, the 3D camera and the hyperspectral camera, the image algorithm processing module performs processing by the following steps:

[0013] S1, matching the height image collected by the 3D camera with the color image collected by the color camera through threshold segmentation and scaling to obtain a color image, and performing image grayscale, image filtering, image binarization, edge detection and contour analysis on the color image in sequence to obtain the center of mass and rotation angle of the material, thereby obtaining the location information of the material;

[0014] S2, converting the RGB model of the color image into an HSV model, equalizing it, removing noise, and then distinguishing dark objects from non-dark objects by detecting colors;

[0015] S3, importing the hyperspectral image information of the non-dark object collected by the hyperspectral camera into the prediction model for classification prediction to obtain category information of the non-dark object;

[0016] S4, selecting three spectral features in the hyperspectral image information to obtain a three-channel pseudo-color image similar to the color image, and then matching the pseudo-color image with the height image after threshold segmentation and scaling to obtain the position information and height features of the non-dark object.

[0017] As an improvement of the present invention, the method for constructing the prediction model is as follows:

[0018] S3.1, extracting 28 effective features from the wavelength features of the hyperspectral image information by wavelet transformation;

[0019] S3.2, importing the hyperspectral image information after completing step S3.1 into the model of the extreme learning machine for training to obtain the trained parameter weights;

[0020] S3.3, put the parameter weights into the pre-built model to obtain the prediction model.

[0021] By adopting the above technical solution, the present invention has the following beneficial effects:

[0022] 1. The present invention collects multi-dimensional feature information of domestic waste by combining 3D cameras, color cameras and hyperspectral cameras, and can significantly improve the recognition accuracy of domestic waste through multi-feature information fusion and hierarchical identification, so that machine intelligent sorting can be applied to domestic waste sorting, which helps to improve sorting efficiency and improve sorting conditions.

[0023] 2. Through layered recognition and multi-feature fusion, the location and category of domestic waste can be accurately and efficiently detected, the accuracy of domestic waste recognition can be improved, and the problem that near-infrared spectral cameras have difficulty in recognizing dark domestic waste and ordinary color cameras have difficulty in recognizing transparent domestic waste can be effectively solved.

[0024] 3. Materials are transported by conveyor belts, and multi-camera image acquisition is used to obtain the material's height information, color information, spatial information, and spectral information. By processing the material's color information, dark-colored materials with very low added value in domestic waste can be screened out, and by processing the spectral information and height information, valuable location and category information of non-dark-colored materials can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a structural schematic diagram of the domestic waste identification system of the present invention;

[0026] Figure 2 It is a schematic diagram of the image algorithm processing flow of the domestic waste identification system of the present invention.

[0027] The markings in the figure correspond to the following:

[0028] 10-Garbage conveying device; 20-Color camera;

[0029] 21- linear light source; 30- 3D camera;

[0030] 31-Laser transmitter; 40-Hyperspectral camera;

[0031] 41-Dome light source; 42-Halogen lamp. DETAILED DESCRIPTION

[0032] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0033] like Figure 1As shown, this embodiment provides a domestic waste identification system, including an industrial computer (not shown in the figure), a garbage conveying device 10, and a color camera 20, a 3D camera 30 and a hyperspectral camera 40 located above the garbage conveying device 10 and arranged in sequence along the conveying direction of the garbage conveying device 10, wherein the industrial computer has an image algorithm processing module, which is mainly used to perform multi-feature fusion on image information collected by multiple cameras and combine the algorithm to obtain the location information and category of the material. This type of industrial computer with an image algorithm processing module can be purchased directly from the market. Of course, when necessary, it is necessary to install software on the industrial computer to realize the required control processing function.

[0034] The garbage conveying device 10 is used to transport domestic garbage as a material. When in use, the garbage conveying device 10 is provided with a hopper for dumping garbage at the head end, and a sorting device or a robot for sorting garbage at the tail end (the hopper and the sorting device or the robot are not part of this embodiment). The garbage conveying device 10 can be a conventional belt conveying device, which, together with the color camera 20, the 3D camera 30, the hyperspectral camera 40, and the linear light source 21, the laser transmitter 31, the dome light source 41, and the halogen lamp 42 mentioned below, are all products that can be purchased directly from the market.

[0035] The shooting direction of the color camera 20 is vertically downward, and is used to collect color information and spatial information of the material. A linear light source 21 is arranged at the lower side of the color camera 20 to cooperate with the color camera 20. The irradiation range of the linear light source 21 includes the shooting range of the color camera on the conveying surface of the garbage conveying device 10. The uniform light emitted by the linear light source 21 ensures the stability of the image feature information collection of the color camera 20.

[0036] The 3D camera 30 is used to collect height information of the material. A laser emitter 31 is arranged next to the 3D camera 30. Specifically, the 3D camera and the laser emitter 31 are respectively installed on the same horizontally arranged connecting plate. The light beam of the laser emitter 31 is emitted vertically and is located within the shooting range of the 3D camera 30 on the conveying surface of the garbage conveying device 10. When in use, the 3D camera 30 obtains height profile information through the laser emitted back by the material.

[0037] The hyperspectral camera 40 is used to collect the spectral information and spatial information of the material. A dome light source 41 is arranged directly below the hyperspectral camera 40. The dome light source 41 is preferably a round dome light source, and a halogen lamp 42 is installed on the dome light source 41. A through hole is opened at the center of the dome light source 41. The hyperspectral camera 40 is located directly above the through hole and its shooting direction is vertically downward. In this way, the stability of the material reflection spectrum collection can be ensured by diffuse reflection.

[0038] The specific fixed connection structure of the color camera 20, 3D camera 30 and hyperspectral camera 40 as well as the linear light source 21, the laser transmitter 31 dome light source 41 and the halogen lamp 42 can be a conventional structure, for example, a fixed connection is achieved by a support frame fixedly connected to the frame of the garbage conveying device 10, or a fixed connection is made to a support frame fixed relative to the position of the garbage conveying device 10, which will not be described in detail here. In addition, the garbage conveying device 10, the color camera 20, the 3D camera 30 and the hyperspectral camera 40 are respectively connected to the industrial computer for communication so as to receive and send signals or information, and the specific communication connection structure can be a conventional structure, such as a wire connection or a Bluetooth wireless connection.

[0039] Preferably, the garbage conveying device 10 is provided with a dispersing device (not shown in the figure) for spreading the garbage on the garbage conveying device 10 to prevent the garbage from piling up and affecting the shooting. The dispersing device is located on the side of the color camera 20 away from the 3D camera 30, which can be a commonly used dispersing device in the field of garbage disposal, which is not the focus of this embodiment and will not be described in detail here.

[0040] When in use, the image algorithm processing module screens out dark materials of domestic waste through the image information provided by the color camera 20 and the 3D camera 30, and obtains the location and category information of non-dark materials through the image information provided by the hyperspectral camera 40 and the 3D camera 30. The specific processing method can be an existing method. Considering that the processing efficiency and accuracy of the existing method are relatively low, preferably, this embodiment provides a method with relatively high processing efficiency and accuracy. For details, see Figure 1 and Figure 2 After receiving the image information sent by the color camera 20, the 3D camera 30 and the hyperspectral camera 40, the image algorithm processing module uses the following steps to implement algorithm processing of the image information collected by multiple cameras as multiple features of the material:

[0041] S1, after threshold segmentation and scaling, the height image collected by the 3D camera 30 is matched with the color image collected by the color phase 20 to obtain a color map, and the color map is successively subjected to image graying, image filtering, image binarization, edge detection and contour analysis to obtain the center of mass and rotation angle of the material, thereby obtaining the location information of the material. Since the detected object has obvious height characteristics, and the interference objects in the detection environment are dust impurities with almost no thickness and unclear height characteristics, these interference objects can be well filtered out by the height image, and then, the color map matched with it can also well filter out these interference objects.

[0042] S2, converts the RGB model (i.e., the red, green, and blue primary color model) of the color image into the HSV model (hue, saturation, and brightness model), equalizes it, removes noise, and then distinguishes dark objects from non-dark objects by detecting colors. Dark objects are usually not sorted but discarded directly because their added value is usually relatively low, so they are not further identified.

[0043] S3, importing the hyperspectral image information of the non-dark objects collected by the hyperspectral camera 40 into the prediction model for classification prediction to obtain the category information of the non-dark objects, wherein the prediction model is constructed as follows:

[0044] S3.1, extract 28 effective features from the wavelength features of the hyperspectral image information through wavelet transform. To predict image information, it is first necessary to train the model to obtain parameter weights. Since the spectral dimension of hyperspectral data has 224 wavelength features, directly training it not only has long data but also is very time-consuming. This embodiment extracts 28 effective features through wavelet transform for training and prediction, effectively avoiding the above problems.

[0045] S3.2, importing the hyperspectral image information after completing step S3.1 (i.e., the hyperspectral image data after feature extraction) into the model of the extreme learning machine for training to obtain the trained parameter weights.

[0046] S3.3, put the parameter weights into the pre-built model to obtain the prediction model.

[0047] After the prediction model is built, it can be applied to online real-time detection. At this time, the collected hyperspectral image information of non-dark objects is imported into the prediction model, and the model can derive the category information of the object.

[0048] S4, select three spectral features in the hyperspectral image information to obtain a three-channel pseudo-color image similar to the color image, and then match the pseudo-color image with the height image after threshold segmentation and scaling (similar to step S1) to obtain the position information and height features of the non-dark object.

[0049] The sorting equipment or manipulator sorts the materials according to the above position information and height characteristics. This embodiment mainly collects rich image information of objects through multiple cameras, including color information, spatial information, height information and hyperspectral information. Through algorithm processing, it can well detect and locate dark objects with low added value, and accurately classify and locate non-dark objects, thereby improving the accuracy of subsequent sorting.

[0050] The present invention is described in detail above with reference to the accompanying drawings, but the embodiments of the present invention are not limited to the above embodiments. Those skilled in the art can make various modifications to the present invention based on the prior art, which all belong to the protection scope of the present invention.

Claims

1. A household waste identification system, characterized in that: It includes an industrial computer, a garbage conveying device, and a color camera, a 3D camera, and a hyperspectral camera located above the garbage conveying device and arranged in sequence along the conveying direction of the garbage conveying device, wherein the garbage conveying device, the color camera, the 3D camera, and the hyperspectral camera are respectively connected to the industrial computer for communication; A linear light source cooperating with the color camera is arranged at the lower side of the color camera, a laser emitter cooperating with the 3D camera is arranged next to the 3D camera, and a dome light source cooperating with the hyperspectral camera is arranged directly below the hyperspectral camera; A halogen lamp is installed on the dome light source; The garbage conveying device is provided with a dispersing device for spreading the garbage on the garbage conveying device, and the dispersing device is located on a side of the color camera away from the 3D camera; The industrial computer has an image algorithm processing module, which screens out dark materials of domestic waste through the image information provided by the color camera and the 3D camera, and obtains the location and category information of non-dark materials through the image information provided by the hyperspectral camera and the 3D camera; After receiving the image information sent by the color camera, the 3D camera and the hyperspectral camera, the image algorithm processing module performs processing by the following steps: S1, matching the height image collected by the 3D camera with the color image collected by the color camera through threshold segmentation and scaling to obtain a color image, and performing image grayscale, image filtering, image binarization, edge detection and contour analysis on the color image in sequence to obtain the center of mass and rotation angle of the material, thereby obtaining the location information of the material; S2, converting the RGB model of the color image into an HSV model, equalizing it, removing noise, and then distinguishing dark objects from non-dark objects by detecting colors; S3, importing the hyperspectral image information of the non-dark object collected by the hyperspectral camera into the prediction model for classification prediction to obtain category information of the non-dark object; S4, selecting three spectral features in the hyperspectral image information to obtain a three-channel pseudo-color image similar to the color image, and then matching the pseudo-color image with the height image after threshold segmentation and scaling to obtain position information and height features of non-dark objects; The prediction model is constructed as follows: S3.1, extracting 28 effective features from the wavelength features of the hyperspectral image information by wavelet transformation; S3.2, importing the hyperspectral image information after completing step S3.1 into the model of the extreme learning machine for training to obtain the trained parameter weights; S3.3, put the parameter weights into the pre-built model to obtain the prediction model.

Images