A method for detecting line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators
By using a cross-scale gradient field sensing operator and a support vector machine model, the spectral-spatial gradient field features of the corridor feature synergy are extracted, which solves the problem of low accuracy in identifying corridor anomalies in frigid regions and achieves a highly efficient anomaly sensing effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGCHUN INST OF TECH
- Filing Date
- 2026-04-15
- Publication Date
- 2026-06-02
AI Technical Summary
Existing methods for detecting line corridor movement have low accuracy in frigid regions, mainly because snow cover causes the loss of spectral and texture differences, making it difficult for traditional methods to segment ground features. Deep learning models also lack sufficient labeled samples, resulting in low accuracy.
A method based on cross-scale gradient field sensing operators is adopted. By using the gradient field convergence characteristic operator GFC-Operator and the anomaly sensing model Model, the spectral-spatial gradient field characteristics of the corridor features are extracted. Anomaly identification is performed using a support vector machine model. The cross-scale gradient field sensing operator CSGF-Operator and the gradient field convergence characteristic operator GFC-Operator are established, and anomaly sensing is performed by combining multi-band remote sensing images.
It improves the accuracy and environmental adaptability of line corridor anomaly identification, and can effectively identify anomalies in frigid regions. It avoids the dependence of traditional methods on target separation and single scale, and significantly improves the identification accuracy.
Smart Images
Figure CN122135214A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of corridor anomaly detection technology, and in particular to a method for detecting corridor anomalies in frigid regions based on cross-scale gradient field sensing operators. Background Technology
[0002] By acquiring and analyzing data on features surrounding power line corridors in frigid environments from remote sensing imagery, the relevant data can effectively characterize the synergistic coupling state between vegetation, conductors, towers, and the background, as well as the degree of snow cover and potential anomaly risks. This characteristic information forms the data foundation for assessing line operation safety, formulating inspection strategies, and providing early warnings of potential risks. Therefore, anomaly analysis of features surrounding power line corridors based on remote sensing imagery has significant scientific value and practical significance. Research on automated identification technology for feature types around power line corridors using remote sensing imagery has become a core technical means for rapidly and accurately acquiring information on corridor anomalies.
[0003] Currently, automatic detection methods for power line corridor anomalies mainly fall into two categories: The first category is target separation-based identification methods. These methods use image segmentation techniques (such as thresholding and edge detection) to separate potential targets from the background in images, and then perform feature analysis and identification on the separated areas. This type of method has a clear logic and can achieve certain results when the contrast between the target and the background is significant. However, in power line corridor scenarios in frigid regions, snow cover smooths out the spectral and textural differences of ground objects, causing the disappearance of surface features and resulting in low accuracy in anomaly detection. For example, the conductors and the background may appear similar in color due to snow, making it difficult to accurately segment different ground objects. The second category is deep learning methods. These methods construct convolutional neural network models to perform end-to-end anomaly detection on power line corridor images. This type of method achieves good recognition results when there are sufficient samples. However, deep learning models heavily rely on large-scale labeled samples for training. In power line corridor scenarios in frigid regions, anomaly events are scarce, and the positive and negative samples are extremely imbalanced, making it difficult to meet the model training requirements, thus leading to low accuracy in anomaly detection. Summary of the Invention
[0004] This invention addresses the problem of low accuracy in identifying line corridor anomalies in existing line corridor anomaly sensing methods by proposing a line corridor anomaly sensing method for frigid regions based on cross-scale gradient field sensing operators.
[0005] A method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators is as follows:
[0006] Acquire the multi-band remote sensing image to be tested, call the gradient field convergence characteristic operator GFC-Operator and the anomaly perception model Model to obtain the anomaly perception map of the multi-band remote sensing image to be tested;
[0007] The anomaly perception model is obtained in the following way:
[0008] Input the sample table TrainSet, call the gradient field convergence characteristic operator GFC-Operator to obtain the gradient field convergence characteristic vector GFCDescVector of the samples in TrainSet, use the sample GFCDescVector and sample state to train the support vector machine classification model, and use the trained support vector machine classification model as the anomaly perception model Model.
[0009] Each record in the TrainSet represents a pixel sample, and each pixel sample includes the following fields: the row (XRow) of the pixel sample in the multi-band remote sensing image, the column (XCol) of the pixel sample in the multi-band remote sensing image, and the state (Decsion) of the pixel sample.
[0010] The pixel sample status Decsion is 0 or 1; 0 indicates that the line corridor has no changes; 1 indicates that the line corridor has changes.
[0011] Furthermore, the gradient field convergence characteristic operator GFC-Operator is obtained in the following way:
[0012] S1. Input a multi-band remote sensing image Image, and use the width and height of the multi-band remote sensing image Image to obtain the multi-scale basis field variable MultiScaleField, the gradient quantization series variable GradLevel, and the field strength tolerance variable FieldTol.
[0013] S2. Establish the cross-scale gradient field sensing operator CSGF-Operator. The input of CSGF-Operator is the row number and column number Col of the pixel in the multi-band remote sensing image Image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale. The output of CSGF-Operator is the gradient field vector CSGFVectorResult.
[0014] S3. Call the operator CSGF-Operator to establish the gradient field convergence characteristic operator GFC-Operator. The input of GFC-Operator is the row number AnaRow and column number AnaCol of the pixel to be analyzed on the multi-band remote sensing image, and the gradient quantization level variable GradLevel. The output is the gradient field convergence characteristic vector GFCCharacterVector.
[0015] Furthermore, in S1, the input multi-band remote sensing image Image is used to obtain the multi-scale basis field variable MultiScaleField, the gradient quantization series variable GradLevel, and the field strength tolerance variable FieldTol, specifically:
[0016] S101. Input multi-band remote sensing image;
[0017] S102. Obtain the width of the multi-band remote sensing image: Width = Image width; Obtain the height of the image: Height = Image height; Obtain the number of bands of the multi-band remote sensing image: Bands = Image number of bands;
[0018] S103. Establish the MultiScaleField, a multi-scale basis phase field variable.
[0019] The multiscale basis phase field variable MultiScaleField is a two-dimensional array with a width of Width and a height of Height, and all elements in the array have a value of 0;
[0020] S104. Initialize the band counter BandCounter=1;
[0021] S105. Temporarily store the contents of the BandCounter band of the single-band image variable TempBandImage=Image;
[0022] S106. Temporarily store the maximum value of a single-band image: TempMax = the maximum value in TempBandImage; TempMin = the minimum value in TempBandImage.
[0023] S107. Let TempBandImage = (TempBandImage - TempMin) / (TempMax - TempMin);
[0024] S108. Let MultiScaleField=MultiScaleField+TempBandImage×(1.5-0.5×BandCounter);
[0025] S109. Set BandCounter = BandCounter + 1;
[0026] S110. If BandCounter > Bands, go to S111; otherwise, go to S105.
[0027] S111. Let MultiScaleField=MultiScaleField / Bands;
[0028] S112. Obtain the mean local variance of the current field strength fluctuation variable FieldOsc=MultiScaleField;
[0029] S113. Obtain the gradient quantization series variable GradLevel using the current field strength fluctuation variable FieldOsc, and output GradLevel.
[0030] S114. Set MultiScaleField = MultiScaleField × GradLevel, and output MultiScaleField;
[0031] S115. Obtain the field strength tolerance variable FieldTol = FieldOsc × 2.2, and output FieldTol.
[0032] Furthermore, in S113, the gradient quantization series variable GradLevel is obtained by using the current field strength fluctuation variable FieldOsc, specifically as follows: GradLevel = round(2.5 / FieldOsc);
[0033] The round() function is used for rounding.
[0034] Furthermore, in S2, a cross-scale gradient field sensing operator CSGF-Operator is established. The inputs of CSGF-Operator are the row and column number (Row and Col) of the pixel in the multi-band remote sensing image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale. The output of CSGF-Operator is the gradient field vector CSGFVectorResult, specifically:
[0035] S201. Establish the cross-scale gradient field sensing operator CSGF-Operator, and input the row number and column number Col of a pixel in a multi-band remote sensing image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale.
[0036] S202, Operator temporary image block SenseBlock = A square with a width of SenseScale is extracted from MultiScaleField with the coordinates composed of Row and Col as the center point;
[0037] S203. Construct the operator temporary storage array CSGFOperatorArray;
[0038] The operator temporary storage array CSGFOperatorArray includes GradLevel elements, and all elements in CSGFOperatorArray have a value of 0;
[0039] S204. Initialize the operator counter CSGFOperatorCounter=1;
[0040] S205, Set the first temporary threshold variable of the operator to CSGFOperatorTemp1 = CSGFOperatorCounter - FieldTol;
[0041] Let the second temporary threshold variable of the operator be CSGFOperatorTemp2 = CSGFOperatorCounter + FieldTol;
[0042] S206, The operator count variable CSGFOperatorTempNum = the sum of the number of pixels in SenseBlock whose values are less than CSGFOperatorTemp1 and greater than CSGFOperatorTemp2;
[0043] S207. Set CSGFOperatorArray[CSGFOperatorCounter] = CSGFOperatorTempNum;
[0044] Wherein, CSGFOperatorArray[CSGFOperatorCounter] is the element with index CSGFOperatorCounter in the operator temporary storage array;
[0045] S208. Set CSGFOperatorCounter = CSGFOperatorCounter + 1;
[0046] S209. If CSGFOperatorCounter is greater than GradLevel, go to S210; otherwise, go to S205.
[0047] S210, The maximum value of the operator temporary storage array CSGFOperatorArrayMax = the maximum value of CSGFOperatorArray;
[0048] S211. Let CSGFOperatorArray = CSGFOperatorArray / (CSGFOperatorArrayMax + 0.01);
[0049] S212. Use CSGFOperatorArray to obtain CSGFVectorResult;
[0050] S213. Output CSGFVectorResult as the result of CSGF-Operator.
[0051] Furthermore, the value of the perception scale variable SenseScale is set in the following way:
[0052] when At that time, ;when At that time, .
[0053] Furthermore, the step of obtaining CSGFVectorResult using CSGFOperatorArray in S212 specifically involves:
[0054] CSGFVectorResult=tanh(CSGFOperatorArray-0.5)×2.
[0055] Furthermore, in S3, the calling operator CSGF-Operator establishes the gradient field convergence characteristic operator GFC-Operator. The inputs of GFC-Operator are the row number AnaRow and column number AnaCol of the pixel to be analyzed in the multi-band remote sensing image, and the gradient quantization level variable GradLevel. The output is the gradient field convergence characteristic vector GFCCharacterVector, specifically:
[0056] S301. Establish the gradient field convergence characteristic operator GFC-Operator, and input the row number AnaRow and column number AnaCol of a pixel to be analyzed in a multi-band remote sensing image.
[0057] S302. Establish the gradient field convergence characteristic vector GFCCharacterVector;
[0058] The gradient field convergence characteristic vector GFCCharacterVector contains GradLevel×3 elements, all of which have a value of 0;
[0059] S303. Establish the gradient memory variable PrevCSGFVector;
[0060] The gradient memory variable PrevCSGFVector contains GradLevel elements, all of which have a value of 0.
[0061] S304. Initialize the multi-scale counter GFCOperatorCounter=1;
[0062] S305, Current Scale = 5 + GFCOperatorCounter × 5;
[0063] S306. Call CSGF-Operator to obtain the current scale gradient vector CurrentCSGFVector;
[0064] Wherein, the input of CSGF-Operator is Row=AnaRow, Col=AnaCol, SenseScale=CurrentSenseScale; the output of CSGF-Operator is CurrentCSGFVector;
[0065] S307, ScaleGradient = CurrentCSGFVector - PrevCSGFVector;
[0066] S308. Let PrevCSGFVector=CurrentCSGFVector;
[0067] S309, the starting position for copying the result vector is StartIndex = (GFCOperatorCounter-1) × GradLevel + 1;
[0068] S310. Copy all contents of ScaleGradient to the elements in GFCCharacterVector located at positions from StartIndex to (StartIndex+GradLevel-1);
[0069] S311. Set GFCOperatorCounter = GFCOperatorCounter + 1;
[0070] S312. If GFCOperatorCounter is greater than 3, go to S313; otherwise, go to S305.
[0071] S313. Output GFCCharacterVector as the result of GFC-Operator.
[0072] Further, the input sample table TrainSet is used to obtain the gradient field convergence characteristic vector GFCDescVector of the samples in TrainSet by calling the gradient field convergence characteristic operator GFC-Operator. The GFCDescVector of the samples and the sample states are then used to train a support vector machine classification model. The trained support vector machine classification model is then used as the anomaly perception model Model. Specifically:
[0073] S401. Input the sample table TrainSet. Each record in TrainSet represents a cell sample, and each cell sample contains the following fields:
[0074] XRow: The row in which a pixel sample is located in the image;
[0075] XCol: The column in the image where the pixel sample is located;
[0076] Decsion: The state of a pixel sample;
[0077] S402. Add a new field, Gradient Field Convergence Characteristic Vector GFCDescVector, to TrainSet. GFCDescVector is a vector that can store GradLevel×3 elements.
[0078] S403, The sample size variable TrainNum = the number of records in TrainSet;
[0079] S404. Initialize the sample counter TrainCounter=1;
[0080] S405, Temporary row TempReco = TrainSet[TrainCounter];
[0081] Wherein, TrainSet[TrainCounter] is the sample with index TrainCounter in TrainSet;
[0082] S406. Call GFC-Operator. The input of GFC-Operator is AnaRow = TempReco.XRow, and AnaCol = TempReco.XCol. The output of GFC-Operator is TempReco.GFCDescVector.
[0083] Where TempReco.XRow is the XRow field in TempReco, TempReco.XCol is the XCol field in TempReco, and TempReco.GFCDescVector is the gradient field convergence characteristic vector of TempReco.
[0084] S407. Set TrainCounter = TrainCounter + 1;
[0085] S408. If TrainCounter > TrainNum, then proceed to S409; otherwise, proceed to S405.
[0086] S409. Construct a support vector machine classification model. The input of the support vector machine classification model is the GFCDescVector field of TrainSet, and the output of the model is the Decsion field of TrainSet.
[0087] S410. Use the GFCDescVector and Decsion fields of TrainSet to train a support vector machine classification model, and use the trained support vector machine classification model as the anomaly perception model.
[0088] Furthermore, the process of acquiring the multi-band remote sensing image to be tested, calling the gradient field convergence characteristic operator GFC-Operator and the anomaly sensing model Model, and obtaining the anomaly sensing map of the multi-band remote sensing image to be tested specifically involves:
[0089] S501. Establish an anomaly perception map;
[0090] The Anomaly Map is a two-dimensional array with a width of 1 and a height of 1, and all elements in the array have a value of 0.
[0091] Where Width1 is the width of the multi-band remote sensing image to be tested, and Height1 is the height of the multi-band remote sensing image to be tested;
[0092] S502. Initialize row counter RowCounter=1;
[0093] S503, Initialize column counter ColCounter=1;
[0094] S504. Call GFC-Operator. The input of GFC-Operator is AnaRow=RowCounter and AnaCol=ColCounter. The output is the perception temporary storage vector SenseVector.
[0095] S505. Input SenseVector into Model to obtain the perception temporary prediction result SensePred;
[0096] S506. Set the value of the cell in the RowCounter row and ColCounter column of AnomalyMap to SensePred.
[0097] S507. Set ColCounter = ColCounter + 1;
[0098] S508. If ColCounter is greater than Width1, go to S509; otherwise, go to S504.
[0099] S509. Set RowCounter = RowCounter + 1;
[0100] S510. If RowCounter is greater than Height1, go to S511; otherwise, go to S503.
[0101] S511. Output AnomalyMap as the final anomaly detection result.
[0102] The beneficial effects of this invention are as follows:
[0103] This invention proposes a method for sensing power line corridor anomalies in frigid regions based on a cross-scale gradient field sensing operator. The invention establishes a cross-scale gradient field sensing operator to describe the spectral-spatial gradient field characteristics of the power line corridor's ground feature synergies at multiple continuous scales. These features are then learned using a support vector machine model. The cross-scale gradient field sensing operator and the anomaly sensing model are used to perceive the safety status of the power line corridor, improving the accuracy of anomaly identification. This invention effectively avoids the dependence of traditional identification techniques on target separation and single-scale methods. By extracting the response change characteristics and spatial distribution patterns of the power line corridor's ground feature synergies in the cross-scale gradient field, it characterizes the overall stability and local anomalies of the spectral-spatial structure in the complex environment of frigid regions, thereby significantly improving the accuracy and environmental adaptability of power line corridor anomaly sensing. Attached Figure Description
[0104] Figure 1 Remote sensing image of a line corridor in a frigid region;
[0105] Figure 2 The image corresponding to the MultiScaleField, a multi-scale basis field variable;
[0106] Figure 3 This is a diagram showing the results of the anomaly detection. Detailed Implementation
[0107] This invention abandons the traditional approach of "identifying vegetation and conductors" and proposes a novel concept: "line corridor and ground feature synergy." It is defined as a unified entity with a holistic spectral and spatial response pattern, formed by the interaction of vegetation, conductors, poles, and background under specific environmental conditions in frigid regions. In remote sensing imagery, this synergy manifests as a "spectral-spatial joint response field" with specific statistical distribution and texture characteristics. Anomalies (such as vegetation encroachment or suspended objects) disrupt the local statistical stationarity of this response field. The invention will now be described in conjunction with specific implementation methods.
[0108] Specific Implementation Method 1: The specific process of this implementation method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators is as follows:
[0109] S1. Input a multi-band remote sensing image (Image). Utilize the width (Width) and height (Height) of the multi-band remote sensing image to obtain the multi-scale basis field variable (MultiScaleField), gradient quantization series variable (GradLevel), and field strength tolerance variable (FieldTol). Specifically:
[0110] S101. Input multi-band remote sensing image;
[0111] S102. Obtain the width of the multi-band remote sensing image: Width = Image width; Obtain the height of the image: Height = Image height; Obtain the number of bands of the multi-band remote sensing image: Bands = Image number of bands;
[0112] S103. Establish a multi-scale basis phase field variable MultiScaleField = Create a two-dimensional array with width Width and height Height, and set all elements in the array to 0.
[0113] S104. Initialize the band counter BandCounter=1;
[0114] S105. Temporarily store the contents of the BandCounter band of the single-band image variable TempBandImage=Image;
[0115] S106. Temporarily store the maximum value of a single-band image: TempMax = the maximum value in TempBandImage; TempMin = the minimum value in TempBandImage.
[0116] S107. Let TempBandImage = (TempBandImage - TempMin) / (TempMax - TempMin);
[0117] The subtraction in this step is an element-by-element subtraction;
[0118] S108. Let MultiScaleField=MultiScaleField+TempBandImage×(1.5-0.5×BandCounter);
[0119] S109. Set BandCounter = BandCounter + 1;
[0120] S110. If BandCounter > Bands, go to S111; otherwise, go to S105.
[0121] S111. Let MultiScaleField=MultiScaleField / Bands;
[0122] S112. Obtain the mean local variance of the current field strength fluctuation variable FieldOsc=MultiScaleField;
[0123] S113. Obtain the gradient quantization level variable GradLevel=round(2.5 / FieldOsc) and output GradLevel;
[0124] Among them, round() is the rounding function;
[0125] S114. Set MultiScaleField = MultiScaleField × GradLevel, and output MultiScaleField;
[0126] S115. Obtain the field strength tolerance variable FieldTol = FieldOsc × 2.2, and output FieldTol.
[0127] S2. Establish the cross-scale gradient field sensing operator CSGF-Operator. The inputs of CSGF-Operator are the row and column number (Row and Col) of the pixel in the multi-band remote sensing image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale. The output of CSGF-Operator is the gradient field vector CSGFVectorResult, specifically:
[0128] S201. Establish the cross-scale gradient field sensing operator CSGF-Operator, and input the row number and column number Col of a pixel in a multi-band remote sensing image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale.
[0129] S202, Operator temporary image block SenseBlock = A square with a width of SenseScale is extracted from MultiScaleField with the coordinates composed of Row and Col as the center point;
[0130] Among them, when At that time, Suitable for high-noise, high-texture, frigid corridor areas; when At that time, Suitable for low-noise, smooth, frigid corridor areas;
[0131] S203. Construct the operator temporary storage array CSGFOperatorArray;
[0132] The operator temporary storage array CSGFOperatorArray includes GradLevel elements, and all elements in CSGFOperatorArray have a value of 0;
[0133] S204. Initialize the operator counter CSGFOperatorCounter=1;
[0134] S205, Set the first temporary threshold variable of the operator to CSGFOperatorTemp1 = CSGFOperatorCounter - FieldTol;
[0135] Let the second temporary threshold variable of the operator be CSGFOperatorTemp2 = CSGFOperatorCounter + FieldTol;
[0136] S206, The operator count variable CSGFOperatorTempNum = the sum of the number of pixels in SenseBlock whose values are less than CSGFOperatorTemp1 and greater than CSGFOperatorTemp2;
[0137] S207. Set CSGFOperatorArray[CSGFOperatorCounter] = CSGFOperatorTempNum;
[0138] Wherein, CSGFOperatorArray[CSGFOperatorCounter] is the element with index CSGFOperatorCounter in the operator temporary storage array;
[0139] S208. Set CSGFOperatorCounter = CSGFOperatorCounter + 1;
[0140] S209. If CSGFOperatorCounter is greater than GradLevel, go to S210; otherwise, go to S205.
[0141] S210, The maximum value of the operator temporary storage array CSGFOperatorArrayMax = the maximum value of CSGFOperatorArray;
[0142] S211. Let CSGFOperatorArray = CSGFOperatorArray / (CSGFOperatorArrayMax + 0.01);
[0143] This step involves performing calculations on each value in the numerical values;
[0144] S212, CSGFVectorResult=tanh(CSGFOperatorArray-0.5)×2;
[0145] S213. Output CSGFVectorResult as the result of CSGF-Operator.
[0146] S3. Call the operator CSGF-Operator to establish the gradient field convergence characteristic operator GFC-Operator. The input of GFC-Operator is the row number AnaRow and column number AnaCol of the pixel to be analyzed in the multi-band remote sensing image, and the gradient quantization level variable GradLevel. The output is the gradient field convergence characteristic vector GFCCharacterVector, specifically:
[0147] S301. Establish the gradient field convergence characteristic operator GFC-Operator, and input the row number AnaRow and column number AnaCol of a pixel to be analyzed in a multi-band remote sensing image.
[0148] S302. Establish the gradient field convergence characteristic vector GFCCharacterVector = Establish a vector containing GradLevel×3 elements, where all element values are 0;
[0149] S303. Establish gradient memory variable PrevCSGFVector = Create a vector containing GradLevel elements, where all element values are 0;
[0150] S304. Initialize the multi-scale counter GFCOperatorCounter=1;
[0151] S305, Current Scale = 5 + GFCOperatorCounter × 5;
[0152] S306. Call CSGF-Operator to obtain the current scale gradient vector CurrentCSGFVector;
[0153] Set the input Row of the CSGF-Operator to AnaRow, Col to AnaCol, and SenseScale to CurrentSenseScale; the output of the CSGF-Operator is CurrentCSGFVector.
[0154] S307, ScaleGradient = CurrentCSGFVector - PrevCSGFVector;
[0155] S308. Let PrevCSGFVector=CurrentCSGFVector;
[0156] S309, the starting position for copying the result vector is StartIndex = (GFCOperatorCounter-1) × GradLevel + 1;
[0157] S310. Copy all contents of ScaleGradient to the elements in GFCCharacterVector located at positions from StartIndex to (StartIndex+GradLevel-1);
[0158] S311. Set GFCOperatorCounter = GFCOperatorCounter + 1;
[0159] S312. If GFCOperatorCounter is greater than 3, go to S313; otherwise, go to S305.
[0160] S313. Output GFCCharacterVector as the result of GFC-Operator.
[0161] S4. Input the sample table TrainSet, call GFC-Operator to obtain the gradient field convergence characteristic vector GFCDescVector of the samples in TrainSet, use the sample GFCDescVector and sample state to train the support vector machine classification model, and use the trained support vector machine classification model as the anomaly perception model Model. Specifically:
[0162] S401. Input the sample table TrainSet. Each record in TrainSet represents a cell sample, and each cell sample contains the following fields:
[0163] XRow: The row in which a pixel sample is located in the image;
[0164] XCol: The column in the image where the pixel sample is located;
[0165] Decsion: The state corresponding to the pixel sample (0: the synergy is stable and the line corridor does not change; 1: the synergy changes and the line corridor changes), which is an integer variable;
[0166] S402. Add a new field, Gradient Field Convergence Characteristic Vector GFCDescVector, to TrainSet. The type of GFCDescVector is a vector that can store GradLevel×3 elements.
[0167] S403, The sample size variable TrainNum = the number of records in TrainSet;
[0168] S404. Initialize the sample counter TrainCounter=1;
[0169] S405, Temporary row TempReco = TrainSet[TrainCounter];
[0170] Wherein, TrainSet[TrainCounter] is the sample with index TrainCounter in TrainSet;
[0171] S406. Call GFC-Operator. The input of GFC-Operator is AnaRow = TempReco.XRow, and AnaCol = TempReco.XCol. The output of GFC-Operator is TempReco.GFCDescVector.
[0172] Where TempReco.XRow is the XRow field in TempReco, TempReco.XCol is the XCol field in TempReco, and TempReco.GFCDescVector is the gradient field convergence characteristic vector of TempReco.
[0173] S407. Set TrainCounter = TrainCounter + 1;
[0174] S408. If TrainCounter > TrainNum, then proceed to S409; otherwise, proceed to S405.
[0175] S409. Construct a support vector machine classification model. The input of the support vector machine classification model is the GFCDescVector field of TrainSet, and the output of the model is the Decsion field of TrainSet.
[0176] S410. Use the GFCDescVector and Decsion fields of TrainSet to train a support vector machine classification model, and use the trained support vector machine classification model as the anomaly perception model.
[0177] S5. Acquire the multi-band remote sensing image to be tested. Call GFC-Operator and Model to obtain the anomaly map of the multi-band remote sensing image to be tested, specifically:
[0178] S501. Create Anomaly Map = Create a two-dimensional array with width 1 and height 1, where all elements in the array have a value of 0;
[0179] Where Width1 is the width of the multi-band remote sensing image to be tested, and Height1 is the height of the multi-band remote sensing image to be tested;
[0180] S502. Initialize row counter RowCounter=1;
[0181] S503, Initialize column counter ColCounter=1;
[0182] S504. Call GFC-Operator. The input of GFC-Operator is AnaRow=RowCounter and AnaCol=ColCounter. The output is the perception temporary storage vector SenseVector.
[0183] S505. Input SenseVector into Model to obtain the perception temporary prediction result SensePred;
[0184] S506. Set the value of the cell in the RowCounter row and ColCounter column of AnomalyMap to SensePred.
[0185] S507. Set ColCounter = ColCounter + 1;
[0186] S508. If ColCounter is greater than Width1, go to S509; otherwise, go to S504.
[0187] S509. Set RowCounter = RowCounter + 1;
[0188] S510. If RowCounter is greater than Height1, go to S511; otherwise, go to S503.
[0189] S511. Output AnomalyMap as the final anomaly detection result.
[0190] Example: To verify the beneficial effects of the present invention, the following experiments were conducted in this example:
[0191] Input a remote sensing image of a frigid region, such as... Figure 1 As shown, the image width Width = 1004 and the image height Height = 1067; the multi-scale basis field variable MultiScaleField, the gradient quantization series variable GradLevel, and the field strength tolerance variable FieldTol were obtained. Among them, MultiScaleField is a two-dimensional array, which was converted into an image as shown... Figure 2 As shown in Table 1, a cross-scale gradient field sensing operator CSGF-Operator and a gradient field convergence characteristic operator GFC-Operator are established. The contents of the input sample table TrainSet are shown in Table 1.
[0192] Table 1
[0193]
[0194] Table 1 shows two states, 0 and 1, with five samples for each state. The TrainSet is used to obtain the motion perception model (Model). Finally, the GFC-Operator and Model are used to obtain the image perception results, such as... Figure 3 As shown, yellow indicates high-probability anomaly areas, and red indicates stable areas. Figure 3 The detection results show that the present invention can accurately detect abnormalities in power line corridors.
Claims
1. A method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators, characterized in that: The specific process of the method is as follows: Acquire the multi-band remote sensing image to be tested, call the gradient field convergence characteristic operator GFC-Operator and the anomaly perception model Model to obtain the anomaly perception map of the multi-band remote sensing image to be tested; The anomaly perception model is obtained in the following way: Input the sample table TrainSet, call the gradient field convergence characteristic operator GFC-Operator to obtain the gradient field convergence characteristic vector GFCDescVector of the samples in TrainSet, use the sample GFCDescVector and sample state to train the support vector machine classification model, and use the trained support vector machine classification model as the anomaly perception model Model. Each record in the TrainSet represents a pixel sample, and each pixel sample includes the following fields: the row (XRow) of the pixel sample in the multi-band remote sensing image, the column (XCol) of the pixel sample in the multi-band remote sensing image, and the state (Decsion) of the pixel sample. The pixel sample status Decsion is 0 or 1; 0 indicates that the line corridor has no changes; 1 indicates that the line corridor has changes.
2. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 1, characterized in that: The gradient field convergence characteristic operator GFC-Operator is obtained in the following way: S1. Input a multi-band remote sensing image Image, and use the width and height of the multi-band remote sensing image Image to obtain the multi-scale basis field variable MultiScaleField, the gradient quantization series variable GradLevel, and the field strength tolerance variable FieldTol. S2. Establish the cross-scale gradient field sensing operator CSGF-Operator. The input of CSGF-Operator is the row number and column number Col of the pixel in the multi-band remote sensing image Image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale. The output of CSGF-Operator is the gradient field vector CSGFVectorResult. S3. Call the operator CSGF-Operator to establish the gradient field convergence characteristic operator GFC-Operator. The input of GFC-Operator is the row number AnaRow and column number AnaCol of the pixel to be analyzed on the multi-band remote sensing image, and the gradient quantization level variable GradLevel. The output is the gradient field convergence characteristic vector GFCCharacterVector.
3. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 2, characterized in that: The input multi-band remote sensing image Image in S1 uses the width and height of the multi-band remote sensing image Image to obtain the multi-scale basis field variable MultiScaleField, the gradient quantization series variable GradLevel, and the field strength tolerance variable FieldTol, specifically: S101. Input multi-band remote sensing image; S102. Obtain the width of the multi-band remote sensing image: Width = Image width; Obtain the height of the image: Height = Image height; Obtain the number of bands of the multi-band remote sensing image: Bands = Image number of bands; S103. Establish the MultiScaleField, a multi-scale basis phase field variable. The multiscale basis phase field variable MultiScaleField is a two-dimensional array with a width of Width and a height of Height, and all elements in the array have a value of 0; S104. Initialize the band counter BandCounter=1; S105. Temporarily store the contents of the BandCounter band of the single-band image variable TempBandImage=Image; S106. Temporarily store the maximum value of a single-band image: TempMax = the maximum value in TempBandImage; TempMin = the minimum value in TempBandImage. S107. Let TempBandImage = (TempBandImage - TempMin) / (TempMax - TempMin); S108. Let MultiScaleField=MultiScaleField+TempBandImage×(1.5-0.5×BandCounter); S109. Set BandCounter = BandCounter + 1; S110. If BandCounter > Bands, go to S111; otherwise, go to S105. S111. Let MultiScaleField=MultiScaleField / Bands; S112. Obtain the mean local variance of the current field strength fluctuation variable FieldOsc=MultiScaleField; S113. Obtain the gradient quantization series variable GradLevel using the current field strength fluctuation variable FieldOsc, and output GradLevel. S114. Set MultiScaleField = MultiScaleField × GradLevel, and output MultiScaleField; S115. Obtain the field strength tolerance variable FieldTol = FieldOsc × 2.2, and output FieldTol.
4. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 3, characterized in that: In S113, the gradient quantization series variable GradLevel is obtained by using the current field strength fluctuation variable FieldOsc. Specifically, GradLevel = round(2.5 / FieldOsc). The round() function is used for rounding.
5. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 4, characterized in that: In step S2, the cross-scale gradient field sensing operator CSGF-Operator is established. The inputs to CSGF-Operator are the row and column number (Row and Col) of the pixel in the multi-band remote sensing image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale. The output of CSGF-Operator is the gradient field vector CSGFVectorResult, specifically: S201. Establish the cross-scale gradient field sensing operator CSGF-Operator, and input the row number and column number Col of a pixel in a multi-band remote sensing image, the basis field variable MultiScaleField, the gradient quantization series variable GradLevel, the field strength tolerance variable FieldTol, and the sensing scale variable SenseScale. S202, Operator temporary image block SenseBlock = A square with a width of SenseScale is extracted from MultiScaleField with the coordinates composed of Row and Col as the center point; S203. Construct the operator temporary storage array CSGFOperatorArray; The operator temporary storage array CSGFOperatorArray includes GradLevel elements, and all elements in CSGFOperatorArray have a value of 0; S204. Initialize the operator counter CSGFOperatorCounter=1; S205, Set the first temporary threshold variable of the operator to CSGFOperatorTemp1 = CSGFOperatorCounter - FieldTol; Let the second temporary threshold variable of the operator be CSGFOperatorTemp2 = CSGFOperatorCounter + FieldTol; S206, The operator count variable CSGFOperatorTempNum = the sum of the number of pixels in SenseBlock whose values are less than CSGFOperatorTemp1 and greater than CSGFOperatorTemp2; S207. Set CSGFOperatorArray[CSGFOperatorCounter] = CSGFOperatorTempNum; Wherein, CSGFOperatorArray[CSGFOperatorCounter] is the element with index CSGFOperatorCounter in the operator temporary storage array; S208. Set CSGFOperatorCounter = CSGFOperatorCounter + 1; S209. If CSGFOperatorCounter is greater than GradLevel, go to S210; otherwise, go to S205. S210, The maximum value of the operator temporary storage array CSGFOperatorArrayMax = the maximum value of CSGFOperatorArray; S211. Let CSGFOperatorArray = CSGFOperatorArray / (CSGFOperatorArrayMax + 0.01); S212. Use CSGFOperatorArray to obtain CSGFVectorResult; S213. Output CSGFVectorResult as the result of CSGF-Operator.
6. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 5, characterized in that: The value of the perception scale variable SenseScale is set in the following way: when At that time, ;when At that time, .
7. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 6, characterized in that: The process of obtaining CSGFVectorResult using CSGFOperatorArray in S212 is as follows: CSGFVectorResult=tanh(CSGFOperatorArray-0.5)×2.
8. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 7, characterized in that: The S3 call operator CSGF-Operator establishes the gradient field convergence characteristic operator GFC-Operator. The input of GFC-Operator is the row number AnaRow and column number AnaCol of the pixel to be analyzed in the multi-band remote sensing image, and the gradient quantization level variable GradLevel. The output is the gradient field convergence characteristic vector GFCCharacterVector, specifically: S301. Establish the gradient field convergence characteristic operator GFC-Operator, and input the row number AnaRow and column number AnaCol of a pixel to be analyzed in a multi-band remote sensing image. S302. Establish the gradient field convergence characteristic vector GFCCharacterVector; The gradient field convergence characteristic vector GFCCharacterVector contains GradLevel×3 elements, all of which have a value of 0; S303. Establish the gradient memory variable PrevCSGFVector; The gradient memory variable PrevCSGFVector contains GradLevel elements, all of which have a value of 0. S304. Initialize the multi-scale counter GFCOperatorCounter=1; S305, Current Scale = 5 + GFCOperatorCounter × 5; S306. Call CSGF-Operator to obtain the current scale gradient vector CurrentCSGFVector; Wherein, the input of CSGF-Operator is Row=AnaRow, Col=AnaCol, SenseScale=CurrentSenseScale; the output of CSGF-Operator is CurrentCSGFVector; S307, ScaleGradient = CurrentCSGFVector - PrevCSGFVector; S308. Let PrevCSGFVector=CurrentCSGFVector; S309, the starting position for copying the result vector is StartIndex = (GFCOperatorCounter-1) × GradLevel + 1; S310. Copy all contents of ScaleGradient to the elements in GFCCharacterVector located at positions from StartIndex to (StartIndex+GradLevel-1); S311. Set GFCOperatorCounter = GFCOperatorCounter + 1; S312. If GFCOperatorCounter is greater than 3, go to S313; otherwise, go to S305. S313. Output GFCCharacterVector as the result of GFC-Operator.
9. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 8, characterized in that: The input sample table TrainSet is used to obtain the gradient field convergence characteristic vector GFCDescVector of the samples in TrainSet by calling the gradient field convergence characteristic operator GFC-Operator. The GFCDescVector and sample states are then used to train a support vector machine classification model. The trained support vector machine classification model is then used as the anomaly perception model Model. Specifically: S401. Input the sample table TrainSet. Each record in TrainSet represents a cell sample, and each cell sample contains the following fields: XRow: The row in which a pixel sample is located in the image; XCol: The column in the image where the pixel sample is located; Decsion: The state of a pixel sample; S402. Add a new field, Gradient Field Convergence Characteristic Vector GFCDescVector, to TrainSet. GFCDescVector is a vector that can store GradLevel×3 elements. S403, The sample size variable TrainNum = the number of records in TrainSet; S404. Initialize the sample counter TrainCounter=1; S405, Temporary row TempReco = TrainSet[TrainCounter]; Wherein, TrainSet[TrainCounter] is the sample with index TrainCounter in TrainSet; S406. Call GFC-Operator. The input of GFC-Operator is AnaRow = TempReco.XRow, and AnaCol = TempReco.XCol. The output of GFC-Operator is TempReco.GFCDescVector. Where TempReco.XRow is the XRow field in TempReco, TempReco.XCol is the XCol field in TempReco, and TempReco.GFCDescVector is the gradient field convergence characteristic vector of TempReco. S407. Set TrainCounter = TrainCounter + 1; S408. If TrainCounter > TrainNum, then proceed to S409; otherwise, proceed to S405. S409. Construct a support vector machine classification model. The input of the support vector machine classification model is the GFCDescVector field of TrainSet, and the output of the model is the Decsion field of TrainSet. S410. Use the GFCDescVector and Decsion fields of TrainSet to train a support vector machine classification model, and use the trained support vector machine classification model as the anomaly perception model.
10. The method for sensing line corridor anomalies in frigid regions based on cross-scale gradient field sensing operators according to claim 9, characterized in that: The process of acquiring the multi-band remote sensing image to be tested involves calling the gradient field convergence characteristic operator GFC-Operator and the anomaly sensing model to obtain the anomaly sensing map of the multi-band remote sensing image to be tested, specifically: S501. Establish an anomaly perception map; The Anomaly Map is a two-dimensional array with a width of 1 and a height of 1, and all elements in the array have a value of 0. Where Width1 is the width of the multi-band remote sensing image to be tested, and Height1 is the height of the multi-band remote sensing image to be tested; S502. Initialize row counter RowCounter=1; S503, Initialize column counter ColCounter=1; S504. Call GFC-Operator. The input of GFC-Operator is AnaRow=RowCounter and AnaCol=ColCounter. The output is the perception temporary storage vector SenseVector. S505. Input SenseVector into Model to obtain the perception temporary prediction result SensePred; S506. Set the value of the cell in the RowCounter row and ColCounter column of AnomalyMap to SensePred. S507. Set ColCounter = ColCounter + 1; S508. If ColCounter is greater than Width1, go to S509; otherwise, go to S504. S509. Set RowCounter = RowCounter + 1; S510. If RowCounter is greater than Height1, go to S511; otherwise, go to S503. S511. Output AnomalyMap as the final anomaly detection result.