Method and device for identifying construction waste in ecological protection red line based on remote sensing image

Abstract

The invention discloses a method and a device for identifying construction waste in an ecological protection red line based on a remote sensing image, and belongs to the technical field of satellite remote sensing. The method comprises the following steps: acquiring a remote sensing image covering an area in a to-be-researched ecological protection red line and preprocessing the remote sensing image; extracting and removing a vegetation region and a water body region according to the preprocessed remote sensing image; converting the remote sensing image after the vegetation area and the water body area are removed into a gray level image, and extracting texture features and gray level features based on the gray level image; fusing the extracted texture features and gray features to obtain fused features; and inputting the fusion features into a trained deep learning model based on a convolutional neural network, carrying out deep feature extraction on the fusion features through the deep learning model, and carrying out classification identification according to the extracted deep features to obtain the categories of the construction waste and the distribution area of each category of construction waste. According to the invention, full-automatic, accurate and rapid construction waste extraction is realized.

Description

Technical Field

[0001] The present invention relates to the field of satellite remote sensing technology, and in particular to a method and device for identifying construction waste within an ecological protection red line based on remote sensing images. Background Art

[0002] Ecological red lines refer to areas within ecological spaces that possess particularly important ecological functions and must be strictly protected. These typically include areas with important ecological functions such as water conservation, biodiversity maintenance, soil and water conservation, windbreaks and sand fixation, and coastal ecological stability, as well as ecologically sensitive and vulnerable areas prone to soil erosion, desertification, and rocky desertification. Ecological red lines are the bottom line and lifeline for ensuring and maintaining ecological security.

[0003] Construction waste refers to all types of waste generated during the construction, renovation, expansion, and demolition of various buildings, pipelines, transportation facilities, and home renovations. This includes waste concrete, slag, waste bricks, mud, discarded materials, and other construction waste, as well as other waste. The frequent illegal dumping of construction waste within ecological protection red lines pollutes and damages land, water resources, the atmosphere, and ecosystems, posing serious environmental risks.

[0004] Traditional methods for identifying construction waste based on remote sensing imagery primarily include manual visual interpretation, human-computer interaction, and multispectral indexing. These methods rely on expert experience and interpretation tools, resulting in low efficiency and high subjectivity. Multispectral indexing has limited adaptability to varying environments and lighting conditions, and its ability to identify non-biological features such as construction waste needs improvement.

[0005] The application of deep learning technology has been a major breakthrough in recent years. By training deep learning models, such as convolutional neural networks (CNNs), it is possible to automatically identify and classify construction waste, significantly improving recognition efficiency and accuracy. However, deep learning models typically require large amounts of labeled data for training, making data acquisition difficult, model training expensive, and requiring high generalization capabilities.

[0006] The ecological protection red line covers a large area and a wide range, and construction waste is scattered and of varying scales. How to provide a high-precision and simpler method for identifying construction waste within the ecological protection red line is an urgent problem that needs to be solved in this field. Summary of the Invention

[0007] In order to solve the above technical problems, the present invention provides a method and device for identifying construction waste within the ecological protection red line based on remote sensing images, which realizes fully automatic, accurate and rapid extraction of construction waste, and provides strong technical support for the management of the ecological protection red line.

[0008] The present invention provides the following technical solutions:

[0009] A method for identifying construction waste within an ecological protection red line based on remote sensing images, the method comprising:

[0010] S1: Acquire remote sensing images covering the area within the ecological protection red line to be studied and perform preprocessing;

[0011] S2: Extract and remove vegetation areas and water areas based on the pre-processed remote sensing images;

[0012] S3: converting the remote sensing image after removing the vegetation area and the water area into a grayscale image, and extracting texture features and grayscale features based on the grayscale image;

[0013] S4: Fuse the extracted texture features and grayscale features to obtain fused features;

[0014] S5: Input the fused features into a trained deep learning model based on a convolutional neural network, perform deep feature extraction on the fused features through the deep learning model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0015] Furthermore, the S1 includes:

[0016] S11: Acquire a remote sensing image covering the area within the ecological protection red line to be studied, wherein the remote sensing image is a multi-source remote sensing image;

[0017] S12: performing radiation correction, geometric correction, orthorectification, and image cropping and stitching on the multi-source remote sensing images;

[0018] S13: Fusing the low-resolution multispectral data and the high-resolution panchromatic data contained in the multi-source remote sensing image to obtain a high-resolution color remote sensing image.

[0019] Furthermore, the S2 includes:

[0020] S21: Calculate the normalized vegetation index and normalized water index of each pixel based on the preprocessed remote sensing image;

[0021] S22: georeferencing the calculated normalized difference vegetation index and normalized difference water index data with the pre-processed remote sensing image;

[0022] S23: comparing the normalized vegetation index and normalized water index of each pixel with the set vegetation index threshold and water index threshold, respectively, and classifying the pixels whose normalized vegetation index and normalized water index are respectively greater than the vegetation index threshold and water index threshold as vegetation areas and water areas, respectively;

[0023] S24: Removing the vegetation area and the water area from the pre-processed remote sensing image respectively.

[0024] Furthermore, the S3 includes:

[0025] S301: converting the remote sensing image after removing vegetation areas and water areas into a grayscale image using a weighted calculation method according to the weights of each band of the remote sensing image;

[0026] S302: For each pixel point of the grayscale image, a neighborhood range of a set scale is set with the pixel point as the center pixel point;

[0027] S303: For each neighborhood range, compare the grayscale value of the central pixel point within the neighborhood range with the grayscale values of the surrounding pixels around the central pixel point. If the grayscale value of the surrounding pixels is greater than the grayscale value of the central pixel point, set the value of the surrounding pixels to 1; otherwise, set the value of the surrounding pixels to 0;

[0028] S304: For a neighborhood range of each central pixel point, arrange the values of each surrounding pixel point in the neighborhood range in a specific order to obtain a binary feature code corresponding to each central pixel point of the grayscale image;

[0029] S305: For each binary feature code corresponding to the central pixel, perform a cyclic right shift on the binary feature code, and find the minimum value of each binary feature code obtained by the cyclic right shift as the rotationally invariant binary feature code;

[0030] S306: Convert the binary feature code and the rotation-invariant binary feature code into decimal data to obtain a decimal code value and a rotation-invariant decimal code value corresponding to each central pixel;

[0031] S307: Calculate the probability of occurrence of the rotation-invariant binary feature code B using the full probability formula ;

[0032]

[0033] is the binary feature code with equal probability corresponding to the rotation-invariant binary feature code B, for The probability of occurrence, for The probability of B appearing under the condition that n is the total number of;

[0034] S308: Calculate the rotation-invariant decimal code value corresponding to each central pixel and the probability The product of is used to obtain the texture feature value of each central pixel point, and the texture feature values of all central pixels constitute the texture feature.

[0035] Furthermore, the remote sensing image after removing vegetation areas and water areas is converted into a grayscale image using the following formula;

[0036]

[0037] Among them, G is the grayscale image, R, G, and B are the red band, green band, and blue band of the remote sensing image, respectively. W1, W2, and W3 are the weights of the set red band, green band, and blue band, respectively.

[0038] Furthermore, the S3 further includes:

[0039] S309: Setting neighborhood ranges of different scales, returning to S303 and repeating the process to obtain texture features at multiple scales;

[0040] S310: performing grayscale histogram statistics on the texture features at the multiple scales to obtain grayscale features at the multiple scales.

[0041] Furthermore, the S4 includes:

[0042] The texture features and grayscale features at the multiple scales are fused in series to obtain the fused features.

[0043] Furthermore, the S5 includes:

[0044] S51: Construct a classification system sample set including construction waste samples;

[0045] S52: Constructing a BP neural network model, wherein the BP neural network model includes an input layer, a hidden layer, and an output layer;

[0046] S53: using the classification system sample set to train the BP neural network model, using a gradient descent method and a back propagation algorithm to continuously adjust the weights and biases of the BP neural network model to minimize output error;

[0047] S54: Inputting the fusion features into the BP neural network model, performing deep feature extraction on the fusion features through the BP neural network model, and performing classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0048] Furthermore, the method further comprises:

[0049] S6: Screen the distribution area of construction waste according to the set area threshold.

[0050] A device for identifying construction waste within an ecological protection red line based on remote sensing images, the device comprising:

[0051] Data acquisition and processing module, used to acquire remote sensing images covering the area within the ecological protection red line to be studied and perform preprocessing;

[0052] The removal module is used to extract and remove vegetation areas and water areas based on the pre-processed remote sensing images;

[0053] A feature extraction module is used to convert the remote sensing image after removing the vegetation area and the water area into a grayscale image, and extract texture features and grayscale features based on the grayscale image;

[0054] Feature fusion module, used to fuse the extracted texture features and grayscale features to obtain fused features;

[0055] The classification module is used to input the fusion features into a trained deep learning model based on a convolutional neural network, perform deep feature extraction on the fusion features through the deep learning model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0056] Furthermore, the data acquisition and processing module includes:

[0057] A data acquisition unit, configured to acquire remote sensing images covering the area within the ecological protection red line to be studied, wherein the remote sensing images are multi-source remote sensing images;

[0058] A first processing unit is used to perform radiation correction, geometric correction, orthorectification, and image cropping and splicing on the multi-source remote sensing image;

[0059] The second processing unit is used to fuse the low-resolution multispectral data and the high-resolution panchromatic data contained in the multi-source remote sensing image to obtain a high-resolution color remote sensing image.

[0060] Furthermore, the rejection module includes:

[0061] An index calculation unit is used to calculate the normalized vegetation index and normalized water index of each pixel based on the preprocessed remote sensing image;

[0062] A georeferencing unit is used to georeference the calculated normalized difference vegetation index and normalized difference water index data with the pre-processed remote sensing image;

[0063] The extraction unit is used to compare the normalized vegetation index and normalized water index of each pixel with the set vegetation index threshold and water index threshold, and classify the pixels whose normalized vegetation index and normalized water index are respectively greater than the vegetation index threshold and the water index threshold as vegetation areas and water areas respectively;

[0064] The removal unit is used to remove the vegetation area and the water area from the pre-processed remote sensing image respectively.

[0065] Furthermore, the feature extraction module includes:

[0066] The grayscale unit is used to convert the remote sensing image after removing the vegetation area and water area into a grayscale image by using a weighted calculation method according to the weight of each band of the remote sensing image;

[0067] A neighborhood setting unit, configured to set a neighborhood range of a set scale for each pixel point of the grayscale image, with the pixel point as the center pixel point;

[0068] a comparing unit, configured to compare, for each neighborhood range, the grayscale value of the central pixel point within the neighborhood range with the grayscale values of the surrounding pixels around the central pixel point; if the grayscale value of the surrounding pixels is greater than the grayscale value of the central pixel point, the values of the surrounding pixels are set to 1; otherwise, the values of the surrounding pixels are set to 0;

[0069] an encoding unit for arranging the values of the surrounding pixels in a neighborhood of each central pixel in a specific order to obtain a binary feature code corresponding to each central pixel of the grayscale image;

[0070] a rotationally invariant encoding unit, configured to encode a binary feature corresponding to each central pixel, perform a cyclic right shift on the binary feature code, and find the minimum value of each binary feature code obtained by the cyclic right shift as the rotationally invariant binary feature code;

[0071] A decimal conversion unit is used to convert the binary feature code and the rotation-invariant binary feature code into decimal data to obtain a decimal code value and a rotation-invariant decimal code value corresponding to each central pixel;

[0072] Probability calculation unit, used to calculate the probability of occurrence of rotation-invariant binary feature code B using the full probability formula ;

[0073]

[0074] is the binary feature code with equal probability corresponding to the rotation-invariant binary feature code B, for The probability of occurrence, for The probability of B appearing under the condition that n is the total number of;

[0075] Texture feature calculation unit, used to calculate the rotation invariant decimal code value corresponding to each central pixel and the probability The product of is used to obtain the texture feature value of each central pixel point, and the texture feature values of all central pixels constitute the texture feature.

[0076] Furthermore, the remote sensing image after removing vegetation areas and water areas is converted into a grayscale image using the following formula;

[0077]

[0078] Among them, G is the grayscale image, R, G, and B are the red band, green band, and blue band of the remote sensing image, respectively. W1, W2, and W3 are the weights of the set red band, green band, and blue band, respectively.

[0079] Furthermore, the feature extraction module also includes:

[0080] A multi-scale unit is used to set neighborhood ranges of different scales, return to the comparison unit and repeat the execution to obtain texture features at multiple scales;

[0081] The grayscale feature extraction unit is used to perform grayscale histogram statistics on the texture features at multiple scales to obtain grayscale features at multiple scales.

[0082] Furthermore, the feature fusion module includes:

[0083] The series connection unit is used to connect the texture features and grayscale features at multiple scales in series to fuse them to obtain the fused features.

[0084] Furthermore, the classification module includes:

[0085] A sample set construction unit, used to construct a classification system sample set including construction waste samples;

[0086] A model building unit, used to build a BP neural network model, wherein the BP neural network model includes an input layer, a hidden layer and an output layer;

[0087] A training unit, configured to train the BP neural network model using the classification system sample set, and continuously adjust the weights and biases of the BP neural network model using a gradient descent method and a back propagation algorithm to minimize output error;

[0088] The classification unit is used to input the fusion features into the BP neural network model, perform deep feature extraction on the fusion features through the BP neural network model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0089] Furthermore, the device further comprises:

[0090] The post-processing module is used to screen the distribution area of construction waste according to the set area threshold.

[0091] The present invention has the following beneficial effects:

[0092] The present invention uses multi-source satellite remote sensing image fusion technology to reduce the impact of environmental changes on identification results. At the same time, by combining multiple feature fusions, it effectively alleviates the limitations of single feature identification, realizes fully automated, accurate and rapid extraction of construction waste, and provides strong technical support for the management of ecological protection red lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 This is a flow chart of the method for identifying construction waste within the ecological protection red line based on remote sensing images of the present invention;

[0094] Figure 2 Schematic diagram of calculation and encoding of texture features of the present invention;

[0095] Figure 3 This is a schematic diagram of the device for identifying construction waste within the ecological protection red line based on remote sensing images of the present invention. DETAILED DESCRIPTION

[0096] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0097] Example 1:

[0098] The embodiment of the present invention provides a method for identifying construction waste within the ecological protection red line based on remote sensing images. Figure 1 As shown, the method includes:

[0099] S1: Obtain remote sensing images covering the area within the ecological protection red line to be studied and perform preprocessing.

[0100] In one example, the preprocessing method includes:

[0101] S11: Acquire remote sensing images covering the area within the ecological protection red line to be studied, where the remote sensing images are multi-source remote sensing images.

[0102] For example, multi-source remote sensing images can include Sentinel-2 satellite data and Jilin-1 remote sensing data.

[0103] S12: Perform radiometric correction, geometric correction, orthorectification, and image cropping and stitching on multi-source remote sensing images.

[0104] Radiometric correction: Unify the radiometric values of images acquired by different sensors to the same standard for accurate subsequent analysis.

[0105] Geometric correction: Eliminate the geometric distortion of the image and make the image coordinates match the actual geographic coordinates.

[0106] Image cropping and stitching: crop images according to the scope of the study area and stitch multiple images into a complete large-area image.

[0107] S13: Fusing the low-resolution multispectral data and the high-resolution panchromatic data contained in the multi-source remote sensing image to obtain a high-resolution color remote sensing image.

[0108] For example, the NNDiffuse Pan Sharpening tool can be used to fuse multispectral data from the Jilin-1 satellite with panchromatic data to create a high-resolution remote sensing image. The NNDiffuse Pan Sharpening tool uses a pan-sharping algorithm to fuse a low-resolution multispectral image with a high-resolution grayscale image to produce a high-resolution color image. This tool not only preserves the color information of the multispectral image but also improves the spatial resolution by enhancing contrast and detail, laying the foundation for the automatic identification of construction waste.

[0109] S2: Extract and remove vegetation areas and water areas based on the preprocessed remote sensing images.

[0110] Since the ecological protection red line is mostly covered with vegetation, has a high vegetation coverage rate, and has a considerable amount of water bodies, masking out grassland areas and water areas will greatly improve the efficiency of the model and facilitate visual interpretation of subsequent model classification effects.

[0111] Specifically, the S2 includes:

[0112] S21: Calculate the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) of each pixel based on the preprocessed remote sensing image.

[0113]

[0114]

[0115] in, 、 and They are the reflectance or brightness values of the pixels in the green, near-infrared, and red bands of the Sentinel-2 image, respectively.

[0116] S22: Geo-reference the calculated normalized difference vegetation index and normalized difference water index data with the pre-processed remote sensing image.

[0117] S23: Compare the normalized vegetation index and normalized water index of each pixel with the set vegetation index threshold and water index threshold, and classify the pixels whose normalized vegetation index and normalized water index are respectively greater than the vegetation index threshold and water index threshold as vegetation areas and water areas, respectively.

[0118] S24: respectively remove vegetation areas and water areas from the pre-processed remote sensing image.

[0119] This method calculates the Normalized Difference Vegetation Index (NDVI) and Normalized Water Index (NDWI) based on Sentinel-2 image data. Reclassification is used to extract areas with low NDVI and NDWI values, and a mask is created. Geomatching of the data removes vegetation coverage and water distribution areas within the ecological protection redline of the original image, thereby refining the study area and improving the efficiency of the model's automatic recognition.

[0120] S3: Convert the remote sensing image after removing vegetation areas and water areas into a grayscale image, and extract texture features and grayscale features based on the grayscale image.

[0121] This step is used to extract classification features, including texture features and grayscale features. The present invention does not limit the specific implementation method of extracting texture features and grayscale features. One method is as follows:

[0122] S301: According to the weights of the respective bands of the remote sensing image, a weighted calculation method is adopted to convert the remote sensing image after removing the vegetation area and the water area into a grayscale image.

[0123] When extracting texture features and grayscale features, the image commonly used is a grayscale image. This is because the texture feature and grayscale feature extraction methods of the present invention are both calculated based on grayscale values. Grayscale images can simplify the calculation process while retaining sufficient information for feature extraction.

[0124] In order to highlight the construction waste area, the present invention uses the weighted average method to convert the remote sensing image into a grayscale image. The specific formula is as follows:

[0125]

[0126] Among them, G is the grayscale image, R, G, and B are the red band, green band, and blue band of the remote sensing image, respectively. W1, W2, and W3 are the weights of the set red band, green band, and blue band, respectively.

[0127] By adjusting the weights, the grayscale values of construction waste areas can be more clearly differentiated from those of other areas, making them easier to extract and identify. The specific weights should be adjusted based on actual data and experimental results, with commonly used weights being 0.3, 0.6, and 0.1.

[0128] S302: For each pixel point of the grayscale image, a neighborhood range of a set scale is set with the pixel point as the center pixel point.

[0129] The scale size is determined by the size of the neighborhood it defines (such as 3×3, 5×5, and 7×7, etc.). If the scale is too large, the data volume is small, but the detail information is lost too much; if the scale is too small, the data volume is too large, and the detail information remains relatively detailed. The present invention can select the scale size according to the acquired spatial resolution and spectral resolution.

[0130] S303: For each neighborhood range, compare the grayscale value of the central pixel point in the neighborhood range with the grayscale values of the surrounding pixels around the central pixel point. If the grayscale value of the surrounding pixel points is greater than the grayscale value of the central pixel point, the value of the surrounding pixel points is set to 1, otherwise the value of the surrounding pixel points is set to 0.

[0131] S304: For a neighborhood range of each central pixel point, arrange the values of each surrounding pixel point in the neighborhood range in a specific order to obtain a binary feature code corresponding to each central pixel point of the grayscale image.

[0132] There are (d×d-1) surrounding pixels in a d×d neighborhood, and after comparison, a (d×d-1)-bit binary number can be generated. Taking the 3×3 scale as an example, the 8 surrounding pixels in the 3×3 neighborhood can generate an 8-bit binary number after comparison. Arrange them in a specific order, such as starting from the upper left corner and arranging them clockwise, to obtain a binary feature code. The N pixels in the neighborhood set corresponding to the grayscale image can form different binary patterns.

[0133] by Figure 2 Taking the 3×3 neighborhood shown as an example, the binary feature code obtained is 11010011.

[0134] S305: For each binary feature code corresponding to the central pixel point, perform a cyclic right shift on the binary feature code, and find the minimum value of each binary feature code obtained by the cyclic right shift as the rotation-invariant binary feature code.

[0135] The aforementioned binary feature code is grayscale invariant, but not rotationally invariant. Image rotation results in a different binary feature code. To address this, the present invention continuously rotates the neighborhood (cyclically shifting the neighborhood rightwards) to obtain a series of binary feature codes. The minimum value is taken as the rotationally invariant binary feature code. This rotationally invariant binary feature code is rotationally invariant and remains unchanged regardless of image rotation.

[0136] Rotation-invariant binary feature coding can eliminate some redundant coding information and reduce data dimensionality, but it may result in loss of effective image information. Experiments have shown that when the neighborhood radius R = 1 and N = 8 (i.e., a 3x3 neighborhood), reducing the feature scale from 256 to 36 dimensions, it can eliminate some redundant coding information while also taking into account the image's rotational invariance, significantly preserving image information while reducing the feature scale.

[0137] S306: Convert the binary feature code and the rotation-invariant binary feature code into decimal data to obtain a decimal code value and a rotation-invariant decimal code value corresponding to each central pixel.

[0138] The converted decimal number is the texture feature code of the d×d neighborhood of the central pixel point, and this code value can be used to reflect the texture structure information of the central pixel and the neighborhood area.

[0139] S307: Calculate the probability of occurrence of the rotation-invariant binary feature code B using the full probability formula .

[0140]

[0141] is the binary feature code with equal probability corresponding to the rotation-invariant binary feature code B, for The probability of occurrence, for The probability of B appearing under the condition that n is The total number of .

[0142] Since the rotation-invariant binary feature code corresponds to multiple code values of the original binary feature code, the present invention adopts a full probability formula to solve the problem that the correlation between histogram features cannot be processed due to the influence of prior probability.

[0143] S308: Calculate the rotation-invariant decimal code value and probability corresponding to each central pixel The product of is used to obtain the texture feature value of each central pixel. The texture feature values of all central pixels constitute the texture feature.

[0144] S309: Setting neighborhood ranges of different scales, returning to S303 and repeating the process, to obtain texture features at multiple scales.

[0145] Image texture reflects the structural characteristics of objects within the image, exhibiting characteristics such as scale, anisotropy, and rhythmicity. This paper characterizes the arrangement and combination of texture primitives within the image at multiple scales, as well as their multi-scale variations, to better capture the image's comprehensive structural characteristics and detailed information, revealing the unique characteristics of images at different receptive fields. By performing multi-scale texture feature extraction, the acquired texture features at various scales are serially fused to obtain a comprehensive texture feature for the image.

[0146] S310: performing grayscale histogram statistics on texture features at multiple scales to obtain grayscale features at multiple scales.

[0147] The statistical characteristics of the overall grayscale distribution of the image, that is, the overall distribution of its grayscale values ​​are obtained from the grayscale image. The pixel values ​​in the grayscale image are grayscale levels of 0 to 255, which describe the significant grayscale attribute characteristics of the image. The grayscale distribution of different types of images has its own characteristics. Therefore, by statistically analyzing the grayscale characteristics of each type of image separately, it is possible to obtain the grayscale attribute information contained in each category of image, which plays a certain role in image classification and recognition.

[0148] When calculating the grayscale distribution characteristics of an image, the probability of occurrence of each grayscale level within each image class is statistically analyzed to obtain the grayscale probability distribution histogram corresponding to each image class. Common image grayscale feature metrics include mean, variance, skewness, kurtosis, energy, and entropy. The above transformation of the original grayscale image yields texture features at multiple scales. A grayscale probability distribution histogram is then generated to reveal the texture structure and grayscale distribution corresponding to each code in the statistical histogram at each scale, as well as the grayscale feature metrics corresponding to each code at each scale. This describes the grayscale distribution of a specific structural feature within a local region of the image. Finally, the resulting local image features at multiple scales are fused in series to yield the multi-scale local grayscale features of the image.

[0149] The present invention extracts multi-scale texture features of remote sensing images and obtains corresponding multi-scale grayscale attribute features. Appropriate multiple scales can be selected according to application requirements and actual effects, and the selected multi-scale texture features and multi-scale grayscale attribute features are processed using feature fusion technology to obtain multi-scale comprehensive texture features and multi-scale comprehensive grayscale features, respectively.

[0150] S4: Fuse the extracted texture features and grayscale features to obtain fused features.

[0151] Specifically, texture features and grayscale features at multiple scales can be fused in series to obtain fused features, which are used as overall features for image classification.

[0152] Serial fusion is the process of sequentially connecting multi-scale grayscale features and multi-scale texture features to form a longer feature vector. This method can fully utilize feature information at different scales and improve the expressiveness of features.

[0153] S5: Input the fused features into the trained deep learning model based on convolutional neural network, perform deep feature extraction on the fused features through the deep learning model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0154] As an implementation, S5 includes:

[0155] S51: Construct a classification system sample set including construction waste samples.

[0156] Deep learning models require training before use, and before training, sample sets for various classification systems must be established. Specifically, through visual interpretation of remote sensing images and field research, the spatial distribution of construction waste and non-construction waste such as buildings, roads, and bare land can be identified in remote sensing images. These categories can then be mapped and annotated pixel by pixel, extracting the fusion features of the patches and constructing a construction waste sample library to facilitate subsequent research.

[0157] S52: Construct a BP neural network model, which includes an input layer, a hidden layer, and an output layer.

[0158] The number of neurons in the input layer corresponds to the number of extracted features, the number of neurons in the hidden layer is adjusted according to the complexity of the problem and experimental results, and the number of neurons in the output layer corresponds to the number of categories of the classification task.

[0159] S53: Use the classification system sample set to train the BP neural network model, adopt the gradient descent method and back propagation algorithm, and continuously adjust the weights and biases of the BP neural network model to minimize the output error.

[0160] Specifically, during the training process, the fused features extracted from the classification system sample set are used as input, and the training model is used to classify and identify construction waste and areas such as buildings, roads, and bare land. A certain amount of labeled remote sensing image samples can be used to train the model, continuously optimizing the model's weight parameters to improve recognition accuracy. However, compared to commonly used deep learning methods, the model established by this invention for extracting suspected construction waste requires fewer samples and has higher classification accuracy. Therefore, this model is suitable for research with a low prior sample size and high classification accuracy requirements.

[0161] During the training process, strategies such as momentum method and adaptive learning rate can be used to improve training efficiency and convergence speed.

[0162] S54: Input the fusion features into the BP neural network model, perform deep feature extraction on the fusion features through the BP neural network model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0163] The present invention can classify remote sensing images into construction waste and non-construction waste categories, and then extract the spatial location attributes of the construction waste, such as obtaining the distribution area, calculating the area, etc.

[0164] S6: Screen the distribution area of construction waste according to the set area threshold.

[0165] Based on experience, construction waste has a certain range of area sizes. Areas that are too large or too small do not meet the characteristics of construction waste. Therefore, this paper sets an area threshold, discards areas that do not meet the area threshold requirements, and selects construction waste patches that meet the research requirements. The specific setting value of the area threshold can be adjusted according to actual needs and is not limited by this invention.

[0166] Finally, the image patches of construction waste categories and their attribute information such as area can be exported.

[0167] The present invention is mainly used in ecological red line protection areas. According to the characteristics of the study area, a method for identifying construction waste within the ecological protection red line based on remote sensing images is provided. By utilizing the technical advantages of multi-source satellite remote sensing monitoring, efficient and accurate monitoring and management of construction waste in the ecological protection red line area is achieved, so as to protect and improve the ecological environment, promote the rational recycling and utilization of resources, and provide technical support for the supervision of the ecological protection red line.

[0168] In addition, the present invention not only has the characteristic of high accuracy of feature classification results, but also can make up for the shortcoming that deep learning requires a large number of samples to ensure classification accuracy.

[0169] In the existing technology, the conventional practice is to directly extract and classify features of remote sensing images (after certain preprocessing) through deep learning models. However, remote sensing images have a large amount of data and a lot of redundant information. Not only is the processing efficiency low, but it also usually requires a large amount of labeled data for training. Data acquisition is difficult, model training costs are high, and the model's generalization ability is required to be high. As a result, it is not suitable for the identification of construction waste within the ecological protection red line, which has a large area, a wide range, and scattered distribution of construction waste of varying scales.

[0170] Compared to existing technologies, this method does not directly use a deep learning model to process remote sensing imagery. Instead, it first extracts texture and grayscale features from the remote sensing imagery and then uses the deep learning model to process the extracted texture and depth features. Compared to directly using a deep learning model, extracting texture and grayscale features reduces redundant information, thereby highlighting key features and enhancing category boundaries, thereby improving feature separability. Lower data dimensionality means that neural networks can adopt simpler structures for feature learning and classification, thereby simplifying the model structure and reducing data dimensionality and complexity. Furthermore, texture and grayscale features are robust to common interference factors in remote sensing imagery, such as lighting variations and atmospheric effects. These features can, to a certain extent, mitigate the impact of these interference factors on classification and recognition, enabling the model to maintain stable performance under various environmental conditions. Due to the reduced data dimensionality and more representative features, the neural network can focus more on learning the underlying patterns in the data during training, rather than overfitting to the noise and details in the training samples. This enables the model to generalize more effectively to new remote sensing imagery samples, enabling better recognition results with a smaller sample size. In short, the number of samples required by the present invention is greatly reduced, which not only reduces the workload of data collection and annotation, but also speeds up model training and deployment, allowing the technology to adapt to new data sets more quickly.

[0171] Furthermore, the present invention integrates multiple classification features by fusing texture and grayscale attributes, utilizing a neural network model for classification, ensuring that the model can accurately and efficiently extract the distribution of construction waste. This multi-feature fusion approach improves recognition accuracy and reduces the potential for missed detections caused by a single feature.

[0172] Furthermore, the present invention integrates multi-source remote sensing data and performs refined processing on the study area, effectively reducing the amount of data and thus improving the model's efficiency. This approach reduces the consumption of computing resources and enables the model to quickly respond to and process large amounts of data.

[0173] In summary, the present invention reduces the impact of environmental changes on recognition results through multi-source satellite remote sensing image fusion technology. At the same time, by combining multiple feature fusions, it effectively alleviates the limitations of single feature recognition, realizes fully automated, accurate and rapid extraction of construction waste, and provides strong technical support for the management of ecological protection red lines.

[0174] Example 2:

[0175] The embodiment of the present invention provides a device for identifying construction waste within the ecological protection red line based on remote sensing images, such as Figure 3 As shown, the device includes:

[0176] The data acquisition and processing module 1 is used to acquire remote sensing images covering the area within the ecological protection red line to be studied and perform preprocessing.

[0177] The elimination module 2 is used to extract and eliminate vegetation areas and water areas based on the pre-processed remote sensing image.

[0178] The feature extraction module 3 is used to convert the remote sensing image after removing the vegetation area and the water area into a grayscale image, and extract texture features and grayscale features based on the grayscale image.

[0179] The feature fusion module 4 is used to fuse the extracted texture features and grayscale features to obtain fused features.

[0180] The classification module 5 is used to input the fusion features into the trained deep learning model based on the convolutional neural network, perform deep feature extraction on the fusion features through the deep learning model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0181] As an improvement to this embodiment, the aforementioned data acquisition and processing module includes:

[0182] The data acquisition unit is used to obtain remote sensing images covering the area within the ecological protection red line to be studied, wherein the remote sensing images are multi-source remote sensing images.

[0183] The first processing unit is used to perform radiation correction, geometric correction, orthorectification, and image cropping and splicing on multi-source remote sensing images.

[0184] The second processing unit is used to fuse the low-resolution multispectral data and the high-resolution panchromatic data contained in the multi-source remote sensing image to obtain a high-resolution color remote sensing image.

[0185] In one example, the aforementioned rejection module includes:

[0186] The index calculation unit is used to calculate the normalized vegetation index and normalized water index of each pixel based on the preprocessed remote sensing image.

[0187] The georeferencing unit is used to georeference the calculated normalized difference vegetation index and normalized difference water index data with the preprocessed remote sensing image.

[0188] The extraction unit is used to compare the normalized vegetation index and normalized water index of each pixel with the set vegetation index threshold and water index threshold, and classify the pixels whose normalized vegetation index and normalized water index are respectively greater than the vegetation index threshold and water index threshold as vegetation areas and water areas.

[0189] The removal unit is used to remove vegetation areas and water areas from the pre-processed remote sensing image.

[0190] In the present invention, the aforementioned feature extraction module specifically includes:

[0191] The grayscale unit is used to convert the remote sensing image after removing the vegetation area and water area into a grayscale image by using a weighted calculation method according to the weight of each band of the remote sensing image.

[0192] The calculation formula is:

[0193] Among them, G is the grayscale image, R, G, and B are the red band, green band, and blue band of the remote sensing image, respectively. W1, W2, and W3 are the weights of the set red band, green band, and blue band, respectively.

[0194] The neighborhood setting unit is used to set a neighborhood range of a set scale for each pixel point of the grayscale image, with the pixel point as the center pixel point.

[0195] The comparison unit is used to compare the grayscale value of the central pixel point in each neighborhood range with the grayscale values of the surrounding pixels around the central pixel point. If the grayscale value of the surrounding pixel point is greater than the grayscale value of the central pixel point, the value of the surrounding pixel point is set to 1, otherwise the value of the surrounding pixel point is set to 0.

[0196] The encoding unit is used to arrange the values of the surrounding pixels in the neighborhood range of each central pixel in a specific order to obtain the binary feature code corresponding to each central pixel of the grayscale image.

[0197] The rotation-invariant encoding unit is used to encode the binary feature corresponding to each central pixel point, cyclically right-shift the binary feature code, and find the minimum value of each binary feature code obtained by the cyclic right-shift as the rotation-invariant binary feature code.

[0198] The decimal conversion unit is used to convert the binary feature code and the rotation-invariant binary feature code into decimal data to obtain the decimal code value and the rotation-invariant decimal code value corresponding to each central pixel point.

[0199] Probability calculation unit, used to calculate the probability of occurrence of rotation-invariant binary feature code B using the full probability formula .

[0200]

[0201] is the binary feature code with equal probability corresponding to the rotation-invariant binary feature code B, for The probability of occurrence, for The probability of B appearing under the condition that n is The total number of .

[0202] Texture feature calculation unit, used to calculate the rotation-invariant decimal code value and probability corresponding to each central pixel point The product of is used to obtain the texture feature value of each central pixel. The texture feature values of all central pixels constitute the texture feature.

[0203] The multi-scale unit is used to set the neighborhood range of different scales, return to the comparison unit and repeat the execution to obtain texture features at multiple scales.

[0204] The grayscale feature extraction unit is used to perform grayscale histogram statistics on texture features at multiple scales to obtain grayscale features at multiple scales.

[0205] Accordingly, the aforementioned feature fusion module includes:

[0206] The series unit is used to fuse texture features and grayscale features at multiple scales in series to obtain fused features.

[0207] Exemplarily, the classification module includes:

[0208] The sample set construction unit is used to construct a classification system sample set including construction waste samples.

[0209] The model building unit is used to build a BP neural network model, which includes an input layer, a hidden layer and an output layer.

[0210] The training unit is used to train the BP neural network model using the classification system sample set, and adopts the gradient descent method and back propagation algorithm to continuously adjust the weights and biases of the BP neural network model to minimize the output error.

[0211] The classification unit is used to input the fusion features into the BP neural network model, perform deep feature extraction on the fusion features through the BP neural network model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

[0212] As another improvement to the embodiment of the present invention, the apparatus of the present invention may further include:

[0213] The post-processing module is used to screen the distribution area of construction waste according to the set area threshold.

[0214] The device provided in the embodiment of the present invention has the same implementation principles and technical effects as the aforementioned method embodiment. For the sake of brevity, any matters not mentioned in the device embodiment can be referred to the corresponding contents in the aforementioned method embodiment 1. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the aforementioned devices and units can all refer to the corresponding processes in the aforementioned method embodiment and will not be repeated here.

[0215] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-mentioned embodiments, ordinary technicians in this field should understand that any technician familiar with this technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed by the present invention, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention.

Claims

1. A method for identifying construction waste within the ecological protection red line based on remote sensing images, characterized in that: The method comprises: S1: Acquire remote sensing images covering the area within the ecological protection red line to be studied and perform preprocessing; S2: Extract and remove vegetation areas and water areas based on the pre-processed remote sensing images; S3: converting the remote sensing image after removing the vegetation area and the water area into a grayscale image, and extracting texture features and grayscale features based on the grayscale image; S4: Fuse the extracted texture features and grayscale features to obtain fused features; S5: Input the fused features into a trained deep learning model based on a convolutional neural network, perform deep feature extraction on the fused features through the deep learning model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

2. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 1 is characterized in that: Said S1 comprises: S11: Acquire a remote sensing image covering the area within the ecological protection red line to be studied, wherein the remote sensing image is a multi-source remote sensing image; S12: performing radiation correction, geometric correction, orthorectification, and image cropping and stitching on the multi-source remote sensing images; S13: Fusing the low-resolution multispectral data and the high-resolution panchromatic data contained in the multi-source remote sensing image to obtain a high-resolution color remote sensing image.

3. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 1 is characterized in that: The S2 includes: S21: Calculate the normalized vegetation index and normalized water index of each pixel based on the preprocessed remote sensing image; S22: georeferencing the calculated normalized difference vegetation index and normalized difference water index data with the pre-processed remote sensing image; S23: comparing the normalized vegetation index and normalized water index of each pixel with the set vegetation index threshold and water index threshold, respectively, and classifying the pixels whose normalized vegetation index and normalized water index are respectively greater than the vegetation index threshold and water index threshold as vegetation areas and water areas, respectively; S24: Removing the vegetation area and the water area from the pre-processed remote sensing image respectively.

4. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 1 is characterized in that: The S3 includes: S301: converting the remote sensing image after removing vegetation areas and water areas into a grayscale image using a weighted calculation method according to the weights of each band of the remote sensing image; S302: For each pixel point of the grayscale image, a neighborhood range of a set scale is set with the pixel point as the center pixel point; S303: For each neighborhood range, compare the grayscale value of the central pixel point within the neighborhood range with the grayscale values of the surrounding pixels around the central pixel point. If the grayscale value of the surrounding pixels is greater than the grayscale value of the central pixel point, set the value of the surrounding pixels to 1; otherwise, set the value of the surrounding pixels to 0; S304: For a neighborhood range of each central pixel point, arrange the values of each surrounding pixel point in the neighborhood range in a specific order to obtain a binary feature code corresponding to each central pixel point of the grayscale image; S305: For each binary feature code corresponding to the central pixel, perform a cyclic right shift on the binary feature code, and find the minimum value of each binary feature code obtained by the cyclic right shift as the rotationally invariant binary feature code; S306: Convert the binary feature code and the rotation-invariant binary feature code into decimal data to obtain a decimal code value and a rotation-invariant decimal code value corresponding to each central pixel; S307: Calculate the probability of occurrence of the rotation-invariant binary feature code B using the full probability formula ; is the binary feature code with equal probability corresponding to the rotation-invariant binary feature code B, for The probability of occurrence, for The probability of B appearing under the condition that n is the total number of; S308: Calculate the rotation-invariant decimal code value corresponding to each central pixel and the probability The product of is used to obtain the texture feature value of each central pixel point, and the texture feature values of all central pixels constitute the texture feature.

5. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 4 is characterized in that: The remote sensing image after removing vegetation areas and water areas is converted into a grayscale image using the following formula; Among them, G is the grayscale image, R, G, and B are the red band, green band, and blue band of the remote sensing image, respectively. W1, W2, and W3 are the weights of the set red band, green band, and blue band, respectively.

6. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 4 is characterized in that: Said S3 further comprises: S309: Setting neighborhood ranges of different scales, returning to S303 and repeating the process to obtain texture features at multiple scales; S310: performing grayscale histogram statistics on the texture features at the multiple scales to obtain grayscale features at the multiple scales.

7. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 6 is characterized in that: The S4 includes: The texture features and grayscale features at the multiple scales are fused in series to obtain the fused features.

8. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to any one of claims 1 to 7, characterized in that: The S5 includes: S51: Construct a classification system sample set including construction waste samples; S52: Constructing a BP neural network model, wherein the BP neural network model includes an input layer, a hidden layer, and an output layer; S53: using the classification system sample set to train the BP neural network model, using a gradient descent method and a back propagation algorithm to continuously adjust the weights and biases of the BP neural network model to minimize output error; S54: Inputting the fusion features into the BP neural network model, performing deep feature extraction on the fusion features through the BP neural network model, and performing classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

9. The method for identifying construction waste within the ecological protection red line based on remote sensing images according to claim 8 is characterized in that: The method further comprises: S6: Screen the distribution area of construction waste according to the set area threshold.

10. A device for identifying construction waste within the ecological protection red line based on remote sensing images, characterized in that: The device comprises: Data acquisition and processing module, used to acquire remote sensing images covering the area within the ecological protection red line to be studied and perform preprocessing; The removal module is used to extract and remove vegetation areas and water areas based on the pre-processed remote sensing images; A feature extraction module is used to convert the remote sensing image after removing the vegetation area and the water area into a grayscale image, and extract texture features and grayscale features based on the grayscale image; Feature fusion module, used to fuse the extracted texture features and grayscale features to obtain fused features; The classification module is used to input the fusion features into a trained deep learning model based on a convolutional neural network, perform deep feature extraction on the fusion features through the deep learning model, and perform classification and identification based on the extracted deep features to obtain the categories of construction waste and the distribution areas of each category of construction waste.

Images