A building solid waste recognition method based on millimeter wave radar synthetic aperture imaging

Abstract

The application relates to a building solid waste recognition method based on millimeter wave radar synthetic aperture imaging. An indoor building solid waste accumulation scene is constructed to simulate single solid waste and random accumulation working conditions; millimeter wave radar is used for scanning and sampling, the random accumulation working condition is adjusted for multiple times, SAR images are obtained through imaging processing; images are augmented, various targets are labeled, a data set is divided into a training set and a verification set through segmentation; the training set is put into a model, pre-training weight and hyperparameter setting are carried out, and the best weight file is obtained through training; the image of the steel bar and the whole brick is obtained by filtering the broken brick and the broken concrete block; semantic segmentation is carried out; based on the radar ranging principle, the real space coordinates of the steel bar and the whole brick are obtained, and the recognition of the two-dimensional size is realized. Based on the difference in electromagnetic scattering characteristics, the steel bar and the whole brick are recognized, and based on the radar ranging principle, the two-dimensional size detection is carried out, which provides technical support for the resource utilization and high value-added consumption of building solid waste.

Description

Technical Field

[0001] This invention relates to a method for identifying construction solid waste, and more particularly to a method for identifying construction solid waste based on millimeter-wave radar synthetic aperture imaging, belonging to the field of construction solid waste recycling and application technology. Background Technology

[0002] Steel bars in construction waste have significant recycling value, and relatively flat bricks can be finely sorted and used as landscaping paving and building facade decoration materials, thus achieving high added value utilization. Currently, the most widely used methods for treating construction waste in China are direct landfill disposal or crushing and reuse. Existing waste sorting systems mostly rely on traditional methods, such as coarse screening, magnetic separation, and air separation to separate lightweight materials, followed by manual sorting. Due to the relatively low technological level of construction waste disposal equipment, the actual capacity and overall resource utilization rate of large-scale production lines are generally not ideal, resulting in low added value for the products and a significant gap between the requirements for fine sorting and resource utilization of construction waste.

[0003] Furthermore, there is currently a lack of methods for classifying and two-dimensionally identifying steel bars and whole bricks in construction waste. Existing construction waste identification methods struggle to accurately determine the size of the waste. In practical applications, construction waste is often heavily covered by dust and netting, making the semantic information extracted by computers from purely optical images rather chaotic, resulting in poor recognition success rates and weak robustness. Compared to traditional optical imaging methods, millimeter-wave radar can selectively image steel bars and whole bricks in accumulated construction waste based on differences in electromagnetic scattering characteristics. It can also more accurately identify the location and size of targets. Moreover, millimeter-wave radar is unaffected by lighting conditions, enabling all-weather, all-day operation.

[0004] Therefore, it is necessary to propose a novel method for identifying construction solid waste based on millimeter-wave radar synthetic aperture imaging (SAR) to locate steel bars and whole bricks in construction solid waste and identify their two-dimensional dimensions, thereby providing new technical support for the refined sorting and resource utilization of construction solid waste. Summary of the Invention

[0005] To address the shortcomings of existing technologies and improve the low added value of construction solid waste resource utilization products, this invention provides a construction solid waste identification method based on millimeter-wave radar synthetic aperture imaging. This method identifies steel bars and bricks based on differences in electromagnetic scattering characteristics and performs two-dimensional size detection based on radar ranging principles, providing technical support for the resource utilization and high-value disposal of construction solid waste.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for identifying construction solid waste based on millimeter-wave radar synthetic aperture imaging, comprising the following steps:

[0007] Step 1: Collect various typical construction solid wastes and construct an indoor construction solid waste accumulation scenario to simulate two working conditions: single solid waste and random accumulation. The various typical construction solid wastes include whole bricks, broken bricks, broken concrete blocks and steel bars.

[0008] Step 2: Use millimeter-wave radar to perform planar grating scanning sampling on various typical construction solid wastes under the single solid waste condition. At the same time, make multiple adjustments to various typical construction solid wastes under the random stacking condition. The adjustments include the quantity, position and placement posture of various typical construction solid wastes. After each adjustment, use millimeter-wave radar to perform planar grating scanning sampling. After sampling for both conditions, perform imaging processing on the collected radar echo signals based on the range-Doppler algorithm to obtain SAR images and set them to JPG format.

[0009] Step 3: Augment the acquired SAR images and use open-source data annotation software to annotate various targets in the SAR images to form a dataset. After each SAR image is annotated, a JSON file is generated and automatically saved. The annotation software converts a JSON file into a JPG original image and a PNG mask image. Then, all the JPG original images and PNG mask images are stored in the corresponding folders to form a segmentation dataset. The segmentation dataset is divided into a training set and a validation set in a 9:1 ratio.

[0010] Step 4: Put the training set obtained in Step 3 into the DeepLabv3+ convolutional neural network model, pre-train the weights and hyperparameters, configure the environment required by the algorithm, train the model to obtain the optimal weight file;

[0011] Step 5: Adjust the contrast and, taking advantage of the differences in the electromagnetic scattering characteristics of construction solid waste, filter out broken bricks and concrete blocks in the SAR image under the normalization of the algorithm, thereby selectively obtaining SAR images of steel bars and whole bricks.

[0012] Step 6: Based on the trained DeepLabv3+ convolutional neural network model, perform semantic segmentation on the SAR image, identify the pixel regions where the steel bars and whole bricks are located in the image, classify the targets in the SAR image pixel by pixel, and label them.

[0013] Step 7: Based on the radar ranging principle, with the center of the radar grating scanning plane as the origin of the coordinate system, the distance between the imaging plane and the radar is the z-coordinate of the target position. With the center of the imaging plane as the origin of the Cartesian coordinate system, the x and y coordinates of each point in the SAR image can be obtained. Thus, the true spatial coordinates of each endpoint of the steel bar and the whole brick can be obtained, realizing the identification of the two-dimensional dimensions of the steel bar and the whole brick in the SAR image.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: the material, surface structure, and curvature of a target have a significant impact on its electromagnetic scattering characteristics, thereby affecting its radar imaging effect. Generally speaking, the imaging effect of metallic targets is better than that of non-metallic targets, the imaging effect of planar targets is better than that of curved targets, and the imaging effect of flat targets is better than that of rough targets. This invention proposes a method for identifying steel bars and whole bricks in construction solid waste and obtaining their two-dimensional dimensions based on millimeter-wave radar synthetic aperture imaging technology, utilizing the differences in electromagnetic scattering characteristics. This achieves selective imaging that traditional identification methods cannot achieve. Semantic segmentation is trained based on the DeepLabv3+ convolutional neural network model to identify the pixel regions where steel bars and whole bricks are located, and two-dimensional dimension detection is performed based on the radar ranging principle. Compared with traditional identification methods, this method is more convenient and more accurate, providing technical support for the resource utilization and high-value-added disposal of construction solid waste. Attached Figure Description

[0015] Figure 1 This is a flowchart of the construction solid waste identification method of the present invention;

[0016] Figure 2 This is a comparative example of the actual object and SAR image under a single solid waste working condition according to the present invention;

[0017] Figure 3 This is a comparative example of the actual object and SAR image under the random stacking condition of the present invention;

[0018] Figure 4 This is a comparison diagram of the actual object and the SAR image under the random stacking condition constructed in the embodiment;

[0019] Figure 5 This is a semantic segmentation result diagram of the SAR image in the embodiment;

[0020] Figure 6 This is a schematic diagram of the spatial coordinates of the target position when identifying two-dimensional dimensions according to the present invention;

[0021] Figure 7 This is a one-dimensional distance image of the construction solid waste in the embodiment;

[0022] Figure 8 These are the planar coordinates of the endpoints of the SAR images of the reinforcing bars and whole bricks in the embodiment. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] like Figures 1-3 As shown, a method for identifying construction solid waste based on millimeter-wave radar synthetic aperture imaging is presented, with a specific process combined with... Figure 1 As shown, it includes the following steps:

[0025] Step 1: Collect various typical construction solid wastes and construct an indoor construction solid waste accumulation scenario to simulate two working conditions: single solid waste and random accumulation. The various typical construction solid wastes include whole bricks, broken bricks, broken concrete blocks and steel bars. Under the random accumulation condition, the number of each of the whole bricks, broken bricks and broken concrete blocks is more than 5, and the number of steel bars is more than 1.

[0026] Step 2: Use millimeter-wave radar to perform planar grating scanning sampling on various typical construction solid wastes under the single solid waste condition. See the example of comparing the actual objects with SAR images under the single solid waste condition. Figure 2 As shown, various typical construction solid wastes under the random stacking condition were adjusted multiple times. These adjustments included the quantity, location, and arrangement of each type of typical construction solid waste. After each adjustment, a reasonable scanning rate and step height were set, and planar grating scanning sampling was performed using millimeter-wave radar. An example comparing the actual objects with SAR images under the random stacking condition is provided. Figure 3 As shown, after sampling under both operating conditions, the collected radar echo signals are processed using the range-Doppler algorithm to obtain SAR images, which are then set to JPG format.

[0027] Step 3: Augment the acquired SAR images based on rotation, scaling, and flipping, and use open-source data annotation software to annotate various targets in the SAR images to form a dataset. After each SAR image is annotated, a JSON file is generated and automatically saved. The annotation software converts a JSON file into a JPG original image and a PNG mask image. Then, all the JPG original images and PNG mask images are stored in the corresponding folders to form a segmentation dataset. The segmentation dataset is divided into a training set and a validation set in a 9:1 ratio. The training set is used to train the convolutional neural network model to extract features of various targets, and the validation set is used to evaluate the segmentation effect of the convolutional neural network model.

[0028] Step 4: Put the training set obtained in Step 3 into the DeepLabv3+ convolutional neural network model, perform pre-training weight and hyperparameter settings, configure the environment required by the algorithm to train the model and obtain the optimal weight file for subsequent semantic segmentation of SAR images.

[0029] Step 5: Adjust the contrast and, taking advantage of the differences in the electromagnetic scattering characteristics of construction solid waste, filter out broken bricks and concrete blocks in the SAR image under the normalization of the algorithm, thereby selectively obtaining SAR images of steel bars and whole bricks. The dynamic range of the radar echo signal intensity taken for imaging is set to [-35,0]dB.

[0030] Step 6: Based on the trained DeepLabv3+ convolutional neural network model, perform semantic segmentation on the SAR image, identify the pixel regions where the steel bars and whole bricks are located in the image, classify the targets in the SAR image pixel by pixel, and label them.

[0031] Step 7: Based on the principle of radar ranging, combined with... Figure 6 As shown, with the center of the radar grating scanning plane as the origin of the coordinate system, the distance between the imaging plane and the radar is the z-coordinate of the target position. With the center of the imaging plane as the origin of the Cartesian coordinate system, the x and y coordinates of each point in the SAR image can be obtained. In this way, the true spatial coordinates of each endpoint of the steel bar and the whole brick can be obtained, thus realizing the identification of the two-dimensional dimensions of the steel bar and the whole brick in the SAR image.

[0032] Example

[0033] Step 1: Collect four typical types of construction solid waste: whole bricks, broken bricks, broken concrete blocks, and steel bars. Construct an indoor construction solid waste accumulation scenario, considering both single solid waste and random accumulation conditions. Specifically, select typical construction solid waste from the demolition site. For the single solid waste condition, select one whole brick, one broken brick, one broken concrete block, and one section of steel bar. Combine physical objects with SAR images. Figure 2 As shown, the random stacking condition used 6 whole bricks, 6 broken bricks, and 6 broken concrete blocks, along with 1 section of steel reinforcement. The actual objects and SAR images were combined. Figure 3 As shown, before imaging, absorbing materials are placed in the radar scanning area behind the building solid waste to absorb clutter and improve imaging quality.

[0034] Step 2: Adjust the quantity, location, and arrangement of various typical construction solid wastes under random stacking conditions. Use millimeter-wave radar for planar grating scanning, and then achieve synthetic aperture imaging based on the range-Doppler algorithm. Specifically, during planar scanning sampling, the sampling frequency is 1kHz, the sampling interval is 4ms, the scanning rate is 200mm / s, the step height is 4mm, and the SAR image format is set to JPG. The stacking method must be adjusted before each sampling. The physical sample and SAR image of the randomly stacked waste after one adjustment are combined. Figure 4 As shown;

[0035] Step 3: The acquired SAR images are augmented using the OpenCV2 toolkit, primarily through rotation, scaling, and flipping. Various targets are then labeled using Labelme to form a dataset. The labeling software automatically generates text files containing target locations and types. The acquired and augmented SAR images are preprocessed by dividing the training and validation sets into a 9:1 ratio, generating files in a specified format and path files required for training the model. Broken bricks, broken concrete blocks, whole bricks, and rebars are named brick-piece, concrete-piece, brick, and rebar, respectively.

[0036] In this embodiment, the TIIWR1843 millimeter-wave sensor manufactured by Texas Instruments (TI) is used to scan construction solid waste, and is paired with the DCA1000EVM data acquisition module to capture and visualize signal data in real time, and then pass it to the program for data processing and subsequent applications.

[0037] Step 4: Put the training set obtained in Step 3 into the DeepLabv3+ convolutional neural network model, train it to obtain the segmentation model, set appropriate pre-training weights and hyperparameters, configure the environment required by the algorithm to train the model and obtain the optimal weight file.

[0038] In this example, the backbone network of the DeepLabv3+ convolutional neural network model is MobileNetV2, with 350 training epochs, a downsampling factor of 8, the Sgd optimizer, and the necessary function modules for the algorithm installed. The GPU model is RTX3090.

[0039] Step 5: Adjust the contrast and, taking advantage of the differences in the electromagnetic scattering characteristics of construction solid waste, filter out broken bricks and concrete blocks in the SAR image under the normalization of the algorithm, thereby selectively obtaining SAR images of steel bars and whole bricks. The dynamic range of the radar echo signal intensity taken for imaging is set to [-35,0]dB.

[0040] Step Six: Based on the trained DeepLabv3+ convolutional neural network model, perform semantic segmentation on the SAR image, identify the pixel regions where the steel bars and whole bricks are located, classify the targets in the SAR image pixel by pixel, label them, and output the recognition results. Figure 5 As shown, the lower area represents bricks and the upper area represents steel bars;

[0041] Step 7: Based on the principle of radar ranging, combined with... Figure 6 As shown, with the center of the radar grating scanning plane as the origin of the coordinate system, the distance between the imaging plane and the radar is the z-coordinate of the target position. With the center of the imaging plane as the origin of the Cartesian coordinate system, the x and y coordinates of each point in the SAR image can be obtained. In this way, the true spatial coordinates of each endpoint of the steel bar and the whole brick can be obtained, thus realizing the identification of the two-dimensional dimensions of the steel bar and the whole brick in the SAR image.

[0042] In this example, a one-dimensional distance image of construction solid waste is combined with... Figure 7 As shown, there are 23 range gates between the target surface and the radar. The actual length represented by each range gate, ΔL, can be calculated to be 3.75 cm. The planar coordinates of the endpoints of the SAR images of the reinforcing bars and the entire brick are combined... Figure 8 As shown, taking the center of the radar grating's scanning plane as the origin, the coordinates of the three vertices of the brick in the image plane are (-64.29, 9.74), (178.8, 9.74), and (178.8, -107.1), respectively. Therefore, the brick's dimensions are approximately 243.09 mm × 116.84 mm. The coordinates of the two endpoints of the reinforcing bar are (-155.5, 99.35) and (238.1, 177.3), respectively, which gives the length of the reinforcing bar segment as 401.24 mm.

[0043] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0044] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for construction and demolition waste identification based on millimeter wave radar synthetic aperture imaging, characterized in that: The method comprises the following steps: Step 1: Collect various typical building solid wastes and build an indoor building solid waste accumulation scene, simulate single solid waste and random accumulation conditions, wherein the various typical building solid wastes include whole bricks, brick fragments, concrete fragments and steel bars; Step 2: Use a millimeter wave radar to perform planar raster scanning sampling on the various typical building solid wastes in the single solid waste condition, and adjust the various typical building solid wastes in the random accumulation condition multiple times, the adjustment including the number, position and placement attitude of the various typical building solid wastes, after each adjustment, also use the millimeter wave radar to perform planar raster scanning sampling, after the sampling of the two conditions is completed, perform imaging processing on the collected radar echo signals based on a range-Doppler algorithm, obtain SAR images and set them as jpg format; Step 3: Augment the collected SAR images, and use an open-source data labeling software to label each target in the SAR images to form a data set, after each SAR image is labeled, a json file is generated and automatically saved, the labeling software converts a json file into an original jpg image and a png mask image, then all the jpg original images and png mask images are stored in corresponding folders to form a segmentation data set, and the segmentation data set is divided into a training set and a verification set according to a 9:1 ratio; Step 4: Put the training set obtained in step 3 into a DeepLabv3+ convolutional neural network model, pre-train the weights and set the hyperparameters, configure the required environment of the algorithm, train the model to obtain the best weight file; Step 5: Adjust the contrast, use the difference in electromagnetic scattering characteristics of building solid wastes, under the processing of algorithm normalization, filter out the brick fragments and concrete fragments in the SAR images, so as to selectively obtain the SAR images of steel bars and whole bricks; Step 6: Based on the trained DeepLabv3+ convolutional neural network model, perform semantic segmentation on the whole brick and steel bar SAR images, identify the pixel regions where the steel bars and whole bricks are located in the images, classify the targets in the whole brick and steel bar SAR images pixel by pixel, and label them; Step 7: Based on the radar ranging principle, taking the center of the scanning plane of the radar raster as the coordinate origin, the distance between the imaging plane and the radar is the z coordinate of the target position, taking the center of the imaging plane as the origin of the plane rectangular coordinate system, the x and y coordinates of each point in the SAR image can be obtained, and then the real space coordinates of each end point of the steel bars and whole bricks can be obtained, so as to realize the identification of the two-dimensional size of the steel bars and whole bricks in the SAR image.

2. The method of claim 1, wherein the method is a method of construction and demolition waste identification based on millimeter wave radar synthetic aperture imaging. In step 1, the number of whole bricks, brick fragments and concrete fragments in the random accumulation condition is more than 5 respectively, and the number of steel bars is more than 1.

3. The method of claim 1, wherein the method is a method of construction and demolition waste identification based on millimeter wave radar synthetic aperture imaging. In step 5, the dynamic range of the radar echo signal strength for imaging is set to [-35, 0] dB.

Images