Recycled aggregate component evaluation method and device based on image detection

Abstract

The invention belongs to the technical field of building solid waste recycling, and discloses a recycled aggregate component evaluation method and device based on image detection, and the method comprises the steps: carrying out the image collection of recycled aggregate, obtaining a recycled aggregate surface information image, forming a data set, carrying out the preprocessing of the data set, and dividing the data set into a training set, a verification set and a test set; building an improved YOLOv7 model, training and verifying the model by using the training set and the verification set, and evaluating the performance of the model by using the test set to obtain a recycled aggregate component evaluation model; and inputting a to-be-detected recycled aggregate surface information image into the recycled aggregate component evaluation model, calculating a recycled aggregate surface mortar area proportion, evaluating the mortar content, and outputting an evaluation result. According to the method, the components of the recycled aggregate can be rapidly detected, graded utilization of the recycled aggregate is realized, a scientific basis is provided for graded cyclic utilization of construction wastes, and green cyclic economy oriented to the construction industry is developed.

Description

Technical Field

[0001] This invention belongs to the field of construction solid waste resource utilization technology, and relates to a method and device for evaluating the composition of recycled aggregates based on image detection. Background Technology

[0002] With the continuous growth of the global population and the rapid development of the construction industry, the demand for aggregates has increased significantly, while the emission of construction solid waste has also shown a sharp upward trend. Currently, most construction solid waste is transported directly to the suburbs or urban peripheries for simple landfill or open dumping without any treatment. This not only wastes land and resources but also pollutes the environment. Therefore, developing a green circular economy for the construction industry has become an urgent task. From the perspective of rational resource utilization and environmental protection, the large-scale application of recycled aggregates in concrete preparation and road construction is a solution with dual benefits.

[0003] Due to the presence of old mortar and the micro-cracks accumulated during the crushing and dismantling process, recycled aggregates are characterized by rough surfaces, numerous sharp edges, high porosity, high water absorption, low bulk density, and high crushing index. This presents numerous challenges in the application of recycled aggregates. Research indicates that recycled aggregates with high mortar content, due to their lower mechanical properties, are suitable for non-load-bearing structural engineering projects such as roadbed filling and site backfilling; while recycled aggregates with low mortar content possess excellent physical and mechanical properties, meeting the engineering requirements of load-bearing structures such as road base courses and recycled concrete. Therefore, rapid assessment of the composition of recycled aggregates can effectively optimize their resource utilization. Furthermore, based on the evaluation results of mechanical properties, the applicability of recycled aggregates in road base courses can be clarified, thereby reducing the need for natural aggregate mining, lowering project costs, and meeting the requirements of sustainable development. Therefore, it is necessary to establish an image detection-based method for evaluating the composition of recycled aggregates to facilitate their classification and use. Summary of the Invention

[0004] The purpose of this invention is to provide an image detection-based method for evaluating the composition of recycled aggregates. By detecting recycled aggregates, the content of mortar can be accurately and conveniently evaluated, thereby facilitating the rational and efficient application of recycled aggregates.

[0005] Another objective of this invention is to provide an image detection-based device for evaluating the composition of recycled aggregates.

[0006] To achieve the above objectives, the technical solution adopted by this invention is as follows:

[0007] A method for evaluating the composition of recycled aggregates based on image detection includes the following steps:

[0008] S1. Image acquisition is performed on the recycled aggregate to obtain surface information images of the recycled aggregate, forming a dataset;

[0009] S2. Preprocess the dataset, and then divide the preprocessed dataset into training set, validation set and test set according to the preset ratio;

[0010] S3. Build an improved YOLOv7 model, use the training set and validation set to train and validate the improved YOLOv7 model, use the test set to evaluate the performance of the improved YOLOv7 model, and obtain the recycled aggregate component evaluation model after the set performance index is reached.

[0011] S4. Input the surface information image of the recycled aggregate to be detected into the recycled aggregate component evaluation model, calculate the mortar area ratio on the surface of the recycled aggregate, evaluate the mortar content, and output the evaluation results.

[0012] As a limitation, in step S1, before acquiring images of the recycled aggregate, the construction waste must be pre-processed to obtain the recycled aggregate;

[0013] The pretreatment process for construction waste is as follows:

[0014] S01. The construction waste is crushed in a closed circuit and screened to obtain filler with a particle size of 4.75~20mm;

[0015] S02. Grind the filler for 1-5 minutes at a speed of 60 r / min to obtain ground particles. After sieving again, the particle size is 5-15 mm.

[0016] S03. After cleaning the product to remove the dust adhering to the surface, dry it to obtain recycled aggregate.

[0017] As a second limitation, in step S2, the dataset is preprocessed, which includes data augmentation and annotation.

[0018] As a further definition, the data augmentation process includes: expanding the dataset by means of random cropping, random 90° rotation, flipping, affine transformation, reducing resolution, and randomly adjusting brightness;

[0019] The annotation process includes: annotating the outline anchor frame of recycled aggregate and the outline anchor frame of mortar in the expanded dataset.

[0020] As a further limitation, in step S2, the preprocessed dataset is divided into a training set, a validation set, and a test set according to a preset ratio of 7:2:1.

[0021] As a third constraint, in step S3, the process of building the improved YOLOv7 model is as follows:

[0022] Based on the original YOLOv7 model, the 3×3 convolutions in the first and second ELAN modules of the Backbone network are replaced with DCNv2 deformable convolutions; the 3×3 convolutions in the third and fourth ELAN modules of the Backbone network are replaced with SAConv switchable dilated convolutions; the convolutions output by the fusion of the second, third, and fourth ELAN modules of the Backbone network are replaced with CBAM attention mechanism modules; a CBAM attention mechanism module is also added to the Head network; and the GIoU loss function in YOLOv7 is replaced with the Wise IoU function.

[0023] Add the formula for calculating the area of ​​the anchor frame to the "detect.py" file in YOLOv7.

[0024] As a further restriction, the expression for DCNv2 deformable convolution is:

[0025]

[0026] in, Let be the learnable offset for the k-th sampling point. The modulated scalar that can be learned at the k-th sampling point The weight of the k-th sampling point is... The current position point on the feature map. This refers to the predefined sampling point offset in the standard convolution kernel. For the input feature map, This represents the number of sampling points in the convolution kernel; To output feature map at location The value at that location.

[0027] As a further clarification, the expression for the Wise IoU function is as follows:

[0028]

[0029] in, It is the square of the straight-line distance between the centers of the anchor frames. It is the square of the diagonal distance between the anchor frame and the maximum encirclement of the anchor frame. To be and Operations separated from the computation graph This is the standard IoU loss.

[0030] The present invention also provides an image detection-based device for evaluating the composition of recycled aggregate, comprising a crusher for closed-circuit crushing of construction waste to obtain filler, a self-grinding drum for grinding the filler, a vibrating feeder for conveying recycled aggregate, an image acquisition device disposed on one side of the vibrating feeder for acquiring images of the recycled aggregate on the vibrating feeder, and a computer for receiving and processing data from the image acquisition device. The computer controls the processing parameters of the crusher and the self-grinding drum, and the image acquisition device includes a camera and a supplementary lighting device.

[0031] As a limitation, the camera is fixed above the vibrating feeder, and the supplementary lighting device is a ring LED light or a strip supplementary light.

[0032] The present invention, by adopting the above-described technical solution, achieves the following technical advancements compared to existing technologies:

[0033] (1) The image detection-based recycled aggregate component evaluation method of the present invention first obtains the surface information image of recycled aggregate, and obtains the recycled aggregate component evaluation model based on the improved YOLOv7 model. The recycled aggregate component evaluation model automatically calculates the mortar area ratio on the surface of recycled aggregate and evaluates the mortar content, providing strong support for the classification and utilization of recycled aggregate, realizing rapid evaluation and accurate grading of recycled aggregate quality, and also having important significance for the promotion and application of construction waste backfilling roadbed engineering; compared with the moisture loss ratio test detection method, the detection efficiency of the present invention is significantly improved, and the detection time can be greatly shortened, providing real-time feedback adjustment for subsequent processes; in addition, the whole process does not cause any physical damage to recycled aggregate, realizing non-destructive rapid quality evaluation;

[0034] (2) The method of the present invention obtains a recycled aggregate component evaluation model through the improved YOLOv7 model, which can more accurately calculate the mortar area ratio on the surface of recycled aggregate, evaluate the mortar content, and improve the pertinence and practicality of the model.

[0035] (3) The method of the present invention collects enough surface information images of recycled aggregates to form a dataset, and performs data augmentation and annotation on the dataset. Through data augmentation, the number of surface information images of recycled aggregates is balanced, and through annotation, recycled aggregates and mortars are identified.

[0036] (4) When building the improved YOLOv7 model in this invention, it is based on the original YOLOv7 model. By replacing the 3×3 convolutions in the first and second ELAN modules of the Backbone network with DCNv2 deformable convolutions, the DCNv2 deformable convolutions can help the network extract texture information of the image better, which is of great significance for distinguishing between recycled aggregate and mortar, which have very different appearances. By replacing the 3×3 convolutions in the third and fourth ELAN modules of the Backbone network with SAConv switchable dilated convolutions, the SAConv switchable dilated convolutions can allow the network to extract features under different receptive fields and capture feature information at different scales. The mortar has a rough appearance, while the recycled aggregate has a smooth appearance. The addition of switchable dilated convolutions in the ONV network allows for a better understanding of the texture and roughness of objects at different scales, which is particularly effective for the identification and classification of recycled aggregates and mortar. By replacing the convolutions output by the fusion of the second, third, and fourth ELAN modules in the Backbone network with CBAM attention mechanism modules, and also adding CBAM attention mechanism modules to the Head network, the model is effective in learning subtle differences in the surface texture and color of recycled aggregates and mortar. Moreover, the CBAM attention mechanism has a relatively simple structure and a relatively small number of parameters, achieving a good balance between performance and detection speed, making the model more feasible in practical applications. By replacing the GIoU loss function in YOLOv7 with the Wise IoU function, the Wise IoU function can provide a wise gradient gain allocation strategy, reducing the competitiveness of high-quality anchor boxes while mitigating the adverse effects of low-quality anchor boxes on gradients. By adding a formula for calculating the anchor box area to the "detect.py" file of YOLOv7, the proportion of mortar area on the surface of recycled aggregates can be obtained from the detected image, and the mortar content can be evaluated based on the proportion of mortar area on the surface of recycled aggregates.

[0037] (5) The image detection-based recycled aggregate component evaluation device of the present invention includes a crusher for closed-circuit crushing of construction waste to obtain filler, a self-grinding drum for grinding filler, a vibrating feeder for conveying recycled aggregate, an image acquisition device set on one side of the vibrating feeder for image acquisition of recycled aggregate on the vibrating feeder, and a computer for receiving and processing data from the image acquisition device. The computer controls the processing parameters of the crusher and the self-grinding drum. The image acquisition device includes a camera and a supplementary lighting device. The device of the present invention realizes rapid evaluation and accurate grading of recycled aggregate quality. In addition, the mortar area ratio on the surface of recycled aggregate can also provide real-time feedback for subsequent crushing, screening and strengthening processes, thereby significantly improving the quality of recycled aggregate, realizing the graded utilization of recycled aggregate, and improving the utilization rate of recycled aggregate in engineering.

[0038] In summary, this invention enables rapid detection of recycled aggregate components and facilitates the graded utilization of recycled aggregates, providing a scientific basis for the graded recycling of construction waste and promoting the development of a green circular economy for the construction industry. Attached Figure Description

[0039] Figure 1 The diagram shows a flowchart of the image detection-based method for evaluating the composition of recycled aggregates in Embodiment 1 of the present invention.

[0040] Figure 2 The diagram shown is a data augmentation illustration from Embodiment 1 of the present invention;

[0041] Figure 3 The diagram shown is a structural diagram of the improved YOLOv7 model in Embodiment 1 of the present invention;

[0042] Figure 4 The diagram shown is a schematic of the sampling points of the standard convolution in the original YOLOv7 model in Embodiment 1 of the present invention;

[0043] Figure 5 The diagram shown is a schematic of the sampling points of the DCNv2 deformable convolution in Embodiment 1 of the present invention;

[0044] Figure 6 The diagram shown is a structural diagram of the SAConv switchable dilated convolution in Embodiment 1 of the present invention.

[0045] Figure 7 The diagram shown is a structural diagram of the CBAM attention mechanism module in Embodiment 1 of the present invention;

[0046] Figure 8 The diagram shown illustrates the calculation principle of the Wise IoU function in Embodiment 1 of the present invention.

[0047] Figure 9 The figure shown is a comparison of the detection results of the improved YOLOv7 model in Embodiment 1 of the present invention and the original YOLOv7 model;

[0048] Figure 10 The figure shown is a comparison of the percentage of surface mortar area of ​​recycled aggregate obtained in Example 1 of the present invention and the percentage of water loss obtained in the comparative example as the mortar content increases.

[0049] Figure 11 The figure shown is a linear fitting graph of the ratio of mortar mass loss to water mass loss in Example 1 of the present invention;

[0050] Figure 12 The figure shown is a linear fitting graph of the mortar quality and the proportion of the surface mortar area of ​​the obtained recycled aggregate in Embodiment 1 of the present invention.

[0051] Figure 13The diagram shown is a block diagram of the image detection-based recycled aggregate component evaluation device in Embodiment 2 of the present invention. Detailed Implementation

[0052] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Example 1: An image detection-based method for evaluating the composition of recycled aggregates

[0054] like Figure 1 As shown, this embodiment is a method for evaluating the components of recycled aggregate based on image detection, including the following steps:

[0055] S0. Pre-treat construction waste to obtain recycled aggregate;

[0056] In this step, the pretreatment process for construction waste is as follows:

[0057] S01. After the construction waste is demolished on site by an excavator, it is poured into a crusher for closed-circuit crushing and screening to obtain filler with a particle size of 4.75~20mm.

[0058] In this step, the construction waste is C30 concrete, and the crusher used is a jaw crusher. The parameters of the jaw crusher include the discharge port gap between the moving jaw and the fixed jaw, the throughput and the eccentric shaft speed. The throughput is controlled by adjusting the conveying speed or vibration frequency of the vibrating feeder.

[0059] S02. Place the filler into a self-grinding drum for grinding. The grinding time is 1~5 min and the grinding speed is 60 r / min. The ground particles are obtained and then sieved to obtain a product with a particle size of 5~15 mm.

[0060] The self-grinding drum used in this step has a diameter of 50cm and a length of 70cm, and is equipped with scrapers inside. The particle size of the sieve in step S01 is 4.75~20mm, which is to select a wider particle size range to improve the yield of the crushed product and prevent waste of raw materials. The second sieve is performed in this step to make the particle size meet the actual use requirements.

[0061] S03. After cleaning the product to remove the dust adhering to the surface, dry it to obtain recycled aggregate.

[0062] S1. Use an image acquisition device to acquire images of recycled aggregate to obtain surface information images of recycled aggregate and form a dataset;

[0063] After acquiring images of the recycled aggregate using an image acquisition device, blurry and duplicate images were removed, resulting in 2090 surface information images of the recycled aggregate. The acquired images had a resolution of 3024×4032 pixels and were in JPG format. In this step, it is strictly necessary to ensure that there is no reuse of construction waste used for the verification experiment and recycled aggregate used to build the dataset. This design ensures the scientific rigor of the verification experiment and improves the generalization ability of the model.

[0064] S2. Preprocess the dataset, including data augmentation and annotation; then divide the preprocessed dataset into training set, validation set and test set according to a preset ratio of 7:2:1.

[0065] like Figure 2 As shown, the data augmentation process in this step includes: using the Python third-party image processing library Albumentations to augment the dataset by performing random cropping, random rotation (90°), flipping, affine amplification, low resolution reduction, and random brightness contrast adjustment; by augmenting the dataset, the diversity of real-world scenes can be simulated, improving the model's generalization ability; finally, the dataset is augmented to 3567 images.

[0066] The annotation process includes: using the annotation software labelimg to annotate the outline anchor frames of recycled aggregates and mortars in the expanded dataset to obtain the preprocessed dataset; the preprocessed dataset is divided into training set, validation set and test set according to 7:2:1, resulting in a training set of 2497 images, a validation set of 713 images and a test set of 357 images.

[0067] S3. Build an improved YOLOv7 model, use the training set and validation set to train and validate the improved YOLOv7 model, use the test set to evaluate the performance of the improved YOLOv7 model, and obtain the recycled aggregate component evaluation model after the set performance index is reached.

[0068] like Figure 3 The diagram shown is a structural diagram of the improved YOLOv7 model. The process of building the improved YOLOv7 model is as follows:

[0069] (1) Based on the original YOLOv7 model, the 3×3 convolutions in the first ELAN module and the second ELAN module of the Backbone network are replaced with DCNv2 deformable convolutions; among them, DCNv2 deformable convolutions can adapt to the irregular shapes of recycled aggregates and mortars, which helps to extract more accurate features in the initial feature extraction stage and improve the expressive power and adaptability of features.

[0070] The contours of recycled aggregate and mortar are irregular. DCNv2 deformable convolution allows the network to adjust the sampling position according to the input information to adapt to the irregular shape of recycled aggregate and mortar. DCNv2 deformable convolution can help the network extract the texture information of the image better, which is of great significance for recycled aggregate and mortar with very different appearances. At the same time, the introduction of DCNv2 deformable convolution also enhances the generalization ability of the model and improves the robustness of the model.

[0071] The expression for DCNv2 deformable convolution is:

[0072]

[0073] in, Let be the learnable offset for the k-th sampling point. It is one of the core components of DCNv2 deformable convolution. It is learned from input features by a single convolutional layer, which allows the sampling points to deviate from a fixed grid, thereby better adapting to the actual shape and pose of the object; The modulated scalar that can be learned at the k-th sampling point This is a major improvement of the deformable convolution in DCNv2, with values ​​between 0 and 1. It is also learned by a single convolutional layer, which is used to adjust the magnitude of the feature at each sampling point; The weight of the k-th sampling point is... The current position point on the feature map. This refers to the predefined sampling point offset in the standard convolution kernel. For the input feature map, This represents the number of sampling points in the convolution kernel; To output feature map at location The value at;

[0074] like Figure 4 and Figure 5The diagram shows the sampling points of the standard convolution in the original YOLOv7 model, specifically the 3×3 convolution in the first or second ELAN module of the Backbone network, and the sampling points of the DCNv2 deformable convolution in this embodiment. As can be seen, the receptive field and sampling location of the DCNv2 deformable convolution are different from those of the 3×3 convolution in the first or second ELAN module. The DCNv2 deformable convolution helps the network better extract texture information from images, which is crucial for distinguishing between regenerated aggregate and mortar, which have significantly different appearances. Simultaneously, the introduction of the DCNv2 deformable convolution enhances the model's generalization ability and improves its robustness.

[0075] (2) Replace the 3×3 convolutions in the third and fourth ELAN modules of the Backbone network with SAConv switchable dilated convolutions; among them, for recycled aggregates and mortars with a wide particle size distribution, SAConv switchable dilated convolutions can capture the local and global features of the image more effectively through different dilation rates, realize multi-scale information fusion, and do not change the size of the feature map or increase the parameters, making the model more perceptive and able to better understand the feature information of objects at different scales;

[0076] like Figure 6 As shown, SAConv switchable dilated convolution has three main components: the SAC component and two global context modules placed before and after the SAC component. The core idea of ​​SAConv switchable dilated convolution is to apply different dilation rates to the same input features and then fuse the results of these different convolutions by generating a switching function corresponding to the input features through average pooling.

[0077] use Convolution operation, where For input, For output, As weight, The hyperparameters for SAConv's switchable dilated convolution;

[0078] The formula for replacing the 3×3 convolution with the SAConv switchable dilated convolution is as follows:

[0079]

[0080] in, This is a switch function used to perform the function of an average pooling layer with a 5×5 kernel and a 1×1 convolutional layer. These are trainable weights;

[0081] SAConv switchable dilated convolution allows the network to extract features in different receptive fields and capture feature information at different scales. Since mortar has a rough surface and recycled aggregate has a smooth surface, the addition of SAConv switchable dilated convolution can enable the network to better understand the texture and roughness of objects at different scales. This is particularly effective for the identification and classification of recycled aggregate and mortar.

[0082] (3) Replace the convolution output of the second, third and fourth ELAN modules in the Backbone network with the CBAM attention mechanism module after fusion; and add the CBAM attention mechanism module to the Head network as well; the CBAM attention mechanism module can improve the network’s sensitivity to key features and improve the model’s representation ability. In addition, with the relatively simple design and low computational complexity of the first and second CBAM attention mechanism modules, the model is more feasible in practical applications.

[0083] like Figure 7 As shown, the CBAM attention mechanism module is a lightweight attention module consisting of two sub-modules: the ChannelAttention Module and the Spatial Attention Module, which focus on features in the channel dimension and spatial dimension, respectively. The ChannelAttention Module focuses on meaningful information in the input image. It feeds the input feature image into a max pooling layer and an average pooling layer, reducing the feature image from C×H×W to C×1×1 size. Then, it feeds it into the Shared MLP module to compress the number of channels to 1 / a times the original number, where 'a' is the reduction ratio. Finally, the two results after passing through the Shared MLP module are summed element-wise and merged to output the feature vector. The calculation formula for the ChannelAttention Module is as follows:

[0084]

[0085] Here, MLP represents the Shared MLP module in the Channel Attention Module. and To perform pooling along the channel dimension separately, Calculate the average value of the locations. Calculate the maximum value. It is the sigmoid activation function. and To share two weight matrices of the MLP, This is the output of global average pooling. This is the output of global max pooling;

[0086] In the Shared MLP module, the number of channels is first compressed, then expanded to the original number of channels. Then, the results of the ReLU activation function are summed and merged element by element. Finally, the sigmoid activation function is used to obtain the final output result. Finally, this result is multiplied by the original image to restore it to the size of C×H×W.

[0087] The Spatial Attention Module focuses on the positional information of the input image. The output of the Channel Attention Module is passed through a max-pooling layer and an average-pooling layer to obtain two feature maps. These two feature maps are then concatenated using a concat layer. A convolutional layer is used to change the number of channels, and a sigmoid activation function is applied to generate a spatial attention map. Finally, the output is multiplied by the original image to revert to a size of C×H×W. The calculation formula for the Spatial Attention Module is as follows:

[0088]

[0089] in, This indicates a convolution operation with a filter size of 7×7;

[0090] For the identification of recycled aggregates and mortar, the CBAM attention mechanism module enables the model to learn features more refinedly, dynamically adjust the importance of each feature in the feature map, and learn subtle differences in the texture and color of the object surface. Compared with other attention mechanisms, the CBAM attention mechanism module has a relatively simple structure and fewer parameters, achieving a good balance between performance and detection speed, making the model more feasible in practical applications.

[0091] (4) Replace the GIoU loss function in YOLOv7 with the Wise IoU function; the recognition accuracy and performance of the YOLOv7 model can be improved by adopting the Wise IoU function; add the formula for calculating the anchor box area to the "detect.py" file of YOLOv7; the formula for calculating the anchor box area is existing technology and will not be described here.

[0092] Since step S2 involves using the labelimg software to annotate the anchor boxes for recycled aggregate and mortar contours on the expanded dataset, the resulting processed dataset inevitably contains low-quality samples and some overlapping anchor boxes. Acronymic factors such as distance and aspect ratio exacerbate the penalty for low-quality samples, thus reducing the model's generalization ability. To address this issue, this step considers various IoU losses: the GIoU function uses a penalty term constructed from the minimum bounding box, the DIoU function uses a penalty term constructed from a distance metric, the CIoU function is based on DIoU but considers aspect ratio, and the Wise IoU function uses "outliers" instead of IoU to evaluate anchor box quality and provides a gradient gain allocation strategy, which can reduce the competitiveness of high-quality anchor boxes and mitigate the adverse effects of low-quality anchor boxes on the gradient. Therefore, the Wise IoU function is chosen for this step.

[0093] like Figure 8 The diagram shown illustrates the calculation principle of the Wise IoU function. The expression of the Wise IoU function is as follows:

[0094]

[0095] in, It is the square of the straight-line distance between the centers of the anchor frames. It is the square of the diagonal distance between the anchor frame and the maximum encirclement of the anchor frame. To be and Operations separated from the computation graph For standard IoU loss;

[0096] In this step, after building the improved YOLOv7 model, the improved YOLOv7 model is trained using the training set, then validated using the validation set, and finally evaluated using the test set.

[0097] Figure 9 The image shows a comparison of the detection results of the improved YOLOv7 model and the original YOLOv7 model in this embodiment of the invention. Compared with the original YOLOv7 model, the improved YOLOv7 model in this step, when validated using the validation set, has a precision of 0.911, a recall of 0.911, an mAP@0.5 of 0.951, and an mAP@0.5:0.95 of 0.853, all of which are better than the original YOLOv7 model. When evaluated using the test set, the improved YOLOv7 model in this step has a recall of 0.917, an mAP@0.5 of 0.917, and an mAP@0.5:0.95 of 0.868, all of which are better than the original YOLOv7 model.

[0098] S4. Input the surface information image of the recycled aggregate to be detected into the recycled aggregate component evaluation model, calculate the mortar area ratio on the surface of the recycled aggregate, evaluate the mortar content, and output the evaluation results.

[0099] Experimental verification:

[0100] I. Comparison Indicators

[0101] To objectively evaluate the image detection-based recycled aggregate component assessment method of this embodiment and verify the effectiveness of the model, this embodiment inputs the surface information image of the recycled aggregate to be detected into the recycled aggregate component assessment model, calculates the ratio of the total mortar area to the total detected area, obtains the mortar area ratio on the recycled aggregate surface, denoted as ΔA_rate, and assesses the mortar content. Here, the total mortar area is the total area of ​​the mortar anchor frames, which is the product of the area of ​​each mortar anchor frame and the number of mortar anchor frames.

[0102] In addition, to verify the effectiveness and accuracy of the ΔA_rate index, the water absorption rate of recycled aggregates was compared, and the results of the ΔA_rate index were analyzed using the water loss ratio as a benchmark.

[0103] The principle of the moisture loss ratio experiment is as follows: the recycled aggregate is placed in a drying oven at 75℃ and dried for 24 hours, and then weighed to obtain the mass. Then, heat it at 475℃ for 5 hours, weigh it again, and obtain the mass. The formula for the required water loss ratio is as follows:

[0104]

[0105] in, This represents the moisture loss ratio of the recycled aggregate. The high water absorption rate of the recycled aggregate is due to the high water absorption rate of the attached mortar. Therefore, the more mortar attached, the higher the mortar content, and the higher the water absorption rate, resulting in a greater moisture loss. This is the theoretical basis for the verification experiment. The amount of mortar attached can be estimated by measuring the moisture loss ratio between 75℃ and 475℃. The drying oven temperature of 75℃, instead of 110℃, was chosen to prevent the loss of bound water in the mortar, and the highest temperature of 475℃ was chosen to avoid decarbonization of carbohydrates.

[0106] II. Comparison Process

[0107] The specific setup for the verification experiment is as follows: Five groups of recycled aggregates with different mass ratios of recycled aggregate and attached mortar were selected for the experiment to verify the consistency between the method of this embodiment and the moisture mass loss ratio experiment.

[0108] First, the C30 concrete was crushed and ground. Then, five samples with a total weight of 300g were screened out by manual sorting. The first group contained 300g of recycled aggregate and 0g of mortar; the second group contained 200g of recycled aggregate and 100g of mortar; the third group contained 150g of recycled aggregate and 150g of mortar; the fourth group contained 100g of recycled aggregate and 200g of mortar; and the fifth group contained 0g of recycled aggregate and 300g of mortar. The first to fifth groups were numbered A to E.

[0109] Each group of samples was photographed using an image acquisition device, and the images were input into the recycled aggregate component evaluation model in this embodiment. The proportion of mortar area on the surface of the recycled aggregate was calculated, the mortar content was evaluated, and the evaluation results were output. Each group of samples was photographed four times, and the average value of the four tests was taken as the final result of this embodiment.

[0110] Then, a moisture loss ratio experiment was conducted on the five groups of samples, and the amount of mortar adhesion was estimated by calculating the moisture loss ratio between 75℃ and 475℃.

[0111] like Figure 10 The figure shows a comparison between the mortar arearatio of the recycled aggregate obtained in this embodiment and the mass loss ratio obtained in the comparative example, showing how the mortar content increases. Figure 10 It can be seen that as the moisture loss ratio increases, the ratio of the surface mortar area of ​​recycled aggregate also gradually increases, and the two are positively correlated. This shows that the recycled aggregate component evaluation model obtained in this embodiment based on the YOLOv7 model is effective for the evaluation of recycled aggregate components.

[0112] Studies have shown that the bound water content in recycled aggregate increases linearly with the content of the attached mortar. Based on this, [further research is needed]. Figure 10 The data in the figure were linearly fitted, and the results are as follows: Figure 11 and Figure 12 As shown.

[0113] Depend on Figure 11 It can be seen that the larger the mass of the mortar, the greater the water loss ratio, R. 2 The coefficient reached 0.98, indicating a good linear relationship between the two, which corresponds to the conclusions in existing studies.

[0114] Depend on Figure 12 It can be seen that as the quality of the mortar increases, the proportion of mortar area on the surface of recycled aggregate also gradually increases, R 2 The value reached 0.96, indicating a good linear relationship between the two. The content of mortar in recycled aggregate can be estimated by using the proportion of mortar area on the surface of recycled aggregate.

[0115] In this embodiment, target identification can be performed on recycled aggregates processed by jaw crushers and autogenous grinding drums with different processing parameters. The surface mortar area ratio of the recycled aggregates can be calculated, and the mortar content can be evaluated. Then, the processing effect of the processing parameters can be judged based on the obtained surface mortar area ratio of the recycled aggregates, and a comparison can be made. The feedback is then used to adjust the jaw crusher and autogenous grinding drum to find more suitable processing parameters for the jaw crusher and autogenous grinding drum.

[0116] For example, by evaluating the composition of recycled aggregate at the outlet of a jaw crusher, the results of the composition evaluation of recycled aggregate produced under different crushing parameters can be used to adjust processing parameters such as the discharge port gap between the moving and fixed jaws, the throughput, and the eccentric shaft speed of the jaw crusher. Similarly, by evaluating the composition of recycled aggregate discharged from an auto-grinding drum, processing parameters such as the scraper length, drum speed, feed rate, and processing time can be adjusted.

[0117] It is important to note that the ultimate goal of this embodiment is not to obtain a theoretically absolutely accurate result, but rather to embed it within the crushing and grinding process to provide a relative indicator of quality change. Therefore, as long as it can provide an effective indicator for feedback adjustment of the process flow and achieve good quality control, even systematic errors, and even a certain rate of false detections and false negatives, are acceptable.

[0118] Example 2: An image detection-based device for evaluating the composition of recycled aggregates

[0119] like Figure 13 As shown, this embodiment is an image detection-based recycled aggregate composition evaluation device, used to implement the image detection-based recycled aggregate composition evaluation method of Embodiment 1. It includes a crusher for closed-circuit crushing of construction waste to obtain filler, a self-grinding drum for grinding the filler, a vibrating feeder for conveying recycled aggregate, an image acquisition device set on one side of the vibrating feeder for image acquisition of the recycled aggregate on the vibrating feeder, and a computer for receiving and processing data from the image acquisition device. The computer also controls the processing parameters of the crusher and the self-grinding drum.

[0120] The image acquisition device includes a camera and a supplementary lighting device. The camera is fixed above the vibrating feeder, and the supplementary lighting device is a ring LED light or a strip supplementary light.

[0121] When using an image acquisition device to acquire images of recycled aggregate, the recycled aggregate is first placed on a vibrating feeder and moved. Then, images of the recycled aggregate are acquired by a camera, and supplementary lighting is provided by a supplementary lighting device.

[0122] In this embodiment, a ring LED light is used for supplementary lighting. The ring LED light only needs to illuminate the field of view of the camera, which saves energy. The ring LED light can adapt to target recognition operations and has low energy consumption. However, it should be noted that the supplementary lighting brightness should not be too high to avoid overexposure and distortion, which may lead to detection errors.

[0123] It should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still modify the technical solutions described in the above embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating a recycled aggregate component based on image detection, characterized by, The method comprises the following steps: S1, image acquisition is performed on the recycled aggregate to obtain a recycled aggregate surface information image and form a data set; S2, the data set is preprocessed, and then the preprocessed data set is divided into a training set, a verification set and a test set according to a preset ratio; S3, an improved YOLOv7 model is built, the training set and the verification set are used to train and verify the improved YOLOv7 model, the performance of the improved YOLOv7 model is evaluated using the test set, and when a preset performance index is reached, a recycled aggregate component evaluation model is obtained; S4, a recycled aggregate surface information image to be detected is input into the recycled aggregate component evaluation model, the area proportion of the recycled aggregate surface mortar is calculated, the content of the mortar is evaluated, and an evaluation result is output.

2. The image detection-based recycled aggregate component evaluation method according to claim 1, characterized by, In step S1, before the image acquisition is performed on the recycled aggregate, the construction waste is preprocessed to obtain the recycled aggregate; The construction waste preprocessing process is as follows: S01, the construction waste is crushed in a closed circuit, and the filler with a particle size of 4.75-20 mm is obtained after screening; S02, the filler is ground for 1-5 min at a grinding speed of 60 r / min, the ground particles are obtained, and the product with a particle size of 5-15 mm is obtained after re-screening; S03, the product is dried after being washed to remove the dust attached to the surface.

3. The image detection-based recycled aggregate component evaluation method according to claim 1, characterized by, In step S2, the preprocessing of the data set comprises data enhancement and labeling.

4. The image detection-based recycled aggregate component evaluation method according to claim 3, characterized by, The data enhancement process comprises expanding the data set in the following modes: random cropping, random rotation by 90°, flipping, affine, reducing resolution and random adjusting brightness; The labeling process comprises recycled aggregate contour anchor frame labeling and mortar contour anchor frame labeling on the expanded data set.

5. The image detection-based recycled aggregate component evaluation method according to claim 3, characterized by, In step S2, the preprocessed data set is divided into the training set, the verification set and the test set according to a preset ratio of 7:2:

1.

6. The image detection-based recycled aggregate component evaluation method according to claim 1, characterized by, In step S3, the process of building the improved YOLOv7 model is as follows: Based on the original YOLOv7 model, the 3x3 convolution in the first ELAN module and the second ELAN module in the Backbone network is replaced by the DCNv2 deformable convolution, the 3x3 convolution in the third ELAN module and the fourth ELAN module in the Backbone network is replaced by the SAConv switchable atrous convolution, the convolution output by the fusion of the second ELAN module, the third ELAN module and the fourth ELAN module in the Backbone network is replaced by the CBAM attention mechanism module, the CBAM attention mechanism module is also added in the Head network, and the GIoU loss function in YOLOv7 is replaced by the Wise IoU function; The formula for calculating the anchor frame area is added in the "detect.py" file of YOLOv7.

7. The image detection-based recycled aggregate component evaluation method according to claim 6, characterized by, The expression of the DCNv2 deformable convolution is as follows: wherein, is a learnable offset for the k-th sample point, is a learnable modulation scalar for the k-th sample point, is a weight for the k-th sample point, is a current position point on the feature map, is a predefined sample point offset in the standard convolution kernel, is an input feature map, is the number of sample points in the convolution kernel; is the value of the output feature map at position .

8. The image detection-based recycled aggregate component evaluation method according to claim 6, characterized by, The expression of the Wise IoU function is as follows: where, is the square of the straight-line distance between the anchor box centers, is the square of the diagonal distance between the anchor box and the maximum bounding box, is the square of the diagonal distance between the anchor box and the maximum bounding box, and operations separated from the computation graph, is the standard IoU loss.

9. An apparatus for evaluating a recycled aggregate component based on image detection, characterized by, The application relates to a device for recycling construction waste, which comprises a crusher for crushing the construction waste in a closed circuit to obtain fillers, a self-grinding drum for grinding the fillers, a vibrating feeder for conveying the recycled aggregate, an image acquisition device arranged on one side of the vibrating feeder and used for acquiring images of the recycled aggregate on the vibrating feeder, and a computer used for receiving and processing data of the image acquisition device, wherein the computer controls processing parameters of the crusher and the self-grinding drum, and the image acquisition device comprises a camera and a light supplementing device.

10. The image detection-based recycled aggregate component evaluation apparatus according to claim 9, characterized by, The camera is fixed above the vibrating feeder, and the light supplementing device is a ring-shaped LED lamp or a strip-shaped light supplementing lamp.

Images