Urban wall spectral information disease detection model establishment method, detection method and system
By constructing a city wall spectral information disease detection model and using spectral information expansion network and feature fusion network, the problem of low accuracy of city wall disease detection in the existing technology is solved, and higher accuracy disease detection is achieved.
Patent Information
- Application Number
- CN202411529508.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-30
- Publication Date
- 2025-03-14
AI Technical Summary
In the prior art, the accuracy of city wall disease detection is low, mainly due to the limited number of bands acquired by mounted multispectral imaging equipment and the insufficient spectral information contained, which affects the detection accuracy.
A method for establishing a disease detection model for city wall spectral information is adopted. By obtaining city wall spectral data sets and color image data sets, input spectral information expansion network after preprocessing, and expanding the spectral dimensions, combining feature fusion networks to build a disease detection model, realizing spectral information expansion from sparse channels to multi-channels.
The accuracy of city wall disease detection is improved, ensuring that the acquired spectral data is closer to the spectral reflection characteristics of real objects, supplementing the missing spectral information in the original data, and achieving accurate pixel-by-pixel prediction.
Smart Images
Figure CN119624862A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of digital image processing and relates to a detection method, in particular to a city wall spectrum information disease detection model establishment method, a detection method and a system. Background Art
[0002] The city wall is a typical symbol of urban civilization in the development of human civilization. However, due to the long history of city wall construction, the existing city walls in China have been damaged to varying degrees due to various factors such as climate change and human activities. Accurate detection of these diseases can timely and clearly reflect the current disease status of the city wall, providing a reliable reference for subsequent governance work. The use of spectral imaging equipment mounted on drones for large-scale spectral data acquisition can greatly improve data acquisition efficiency. This method can quickly obtain spectral images of large-area target scenes in different bands in the city wall disease detection task, and spectral remote sensing technology has the advantage of "image-spectrum integration", which can obtain color and texture information while obtaining spectral features. However, the number of bands that can be obtained by mounted multi-spectral imaging equipment is extremely limited, and the spectral information contained is insufficient, which affects the accuracy of city wall disease detection. Summary of the invention
[0003] In view of the deficiencies in the prior art, the object of the present invention is to provide a method for establishing a city wall spectral information disease detection model, a detection method and a system to solve the technical problem of low accuracy in city wall disease detection in the prior art.
[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve the above problems:
[0005] A method for constructing a city wall spectral information disease detection model specifically comprises the following steps:
[0006] Step 1: obtaining a city wall spectral dataset and a city wall color image dataset, and preprocessing the city wall spectral dataset; the city wall spectral dataset includes spectral data of multiple regions of the city wall, and the city wall color image dataset includes color image data of multiple regions of the city wall;
[0007] Step 2: Input the preprocessed city wall spectral dataset into the spectral information expansion network to expand and enhance the spectral dimension, and obtain the expanded city wall spectral dataset;
[0008] Step 3: construct a disease detection model;
[0009] Step 4: Take the expanded city wall spectral dataset obtained in step 2 and the city wall color image dataset obtained in step 1 as input, take the distribution data of city wall surface disease categories in each area as output, train the disease detection model obtained in step 3, and obtain a trained disease detection model; wherein, the city wall surface disease categories include biological diseases, strong salting-out diseases, and peeling diseases.
[0010] Preferably, the disease detection model includes an input layer, a convolution layer and a feature fusion network connected in sequence; the convolution layer includes parallel 2D convolution and 3D convolution, the 2D convolution is used to extract the color image features of the city wall color image data, and the 3D convolution is used to extract the spectral image features of the city wall spectral data; the feature fusion network includes a fusion module and a classifier connected in sequence, the fusion module is used to concatenate all color image features and all spectral image features to obtain a joint feature representation, and the classifier is used to map the joint feature representation to an output dimension with the same number of classification task categories through a linear transformation, and convert it into a probability distribution, and output the city wall surface disease distribution data.
[0011] Preferably, the classifier includes a pooling layer, a fully connected layer and a Softmax layer connected in sequence.
[0012] Preferably, the preprocessing in step 1 is to perform dark current correction and registration on the city wall spectral data set.
[0013] Preferably, the spectral information expansion network includes a 2D module and a 3D spatial spectral expansion module connected in sequence; the 2D module is used to extract the shallow spatial features of the city wall spectral data and output a feature map; the 3D spatial spectral expansion module is used to restore the spatial resolution of the feature map, supplement the missing spectral information, and output the expanded city wall spectral data.
[0014] Preferably, the 2D module includes a 2D preliminary convolution layer, a first attention unit and a first feature fusion module connected in sequence; the first attention unit includes a plurality of first attention modules connected in parallel, each first attention module includes a first convolution layer, a first activation layer, a second convolution layer, a second activation layer residual block and a first channel attention module connected in sequence, and the input end of the first convolution layer is residually connected to the output end of the second activation layer residual block.
[0015] Preferably, the 3D spatial spectrum expansion module includes sequentially connected Dimensional layer,3D preliminary convolution layer, second attention unit and second feature fusion module; the second attention unit includes multiple second attention modules connected in parallel, each second attention module includes a first 3D convolution, a first PRelu activation layer, a second 3D convolution layer, a second PRelu activation layer, a second channel attention module, a multi-layer perceptron, a weighted layer and a Sigmoid activation layer connected in sequence, and the first 3D convolution input end is residually connected to the second PRelu activation layer output end.
[0016] A city wall spectral information expansion disease detection method comprises the following steps:
[0017] S1, collecting the city wall spectrum data and the city wall color image data, and preprocessing the city wall spectrum data;
[0018] S2, inputting the preprocessed city wall spectrum data into the spectrum information expansion network to expand and enhance the spectrum dimension, and obtaining the expanded city wall spectrum data;
[0019] S3, inputting the city wall color image data and the expanded city wall spectral data into the disease detection model constructed by the city wall spectral information disease detection model construction method disclosed in this application for detection, and obtaining the city wall surface disease category distribution data.
[0020] A city wall surface disease detection system is based on the city wall spectral information disease detection model construction method and the city wall spectral information expansion disease detection method disclosed in the present application, comprising an acquisition module, a spectral information expansion module and a disease detection module; the acquisition module is used to obtain city wall spectral data and city wall color image data, and preprocess the city wall spectral data; the spectral information expansion module is used to expand and enhance the spectral dimension of the preprocessed city wall spectral data to obtain expanded city wall spectral data; the disease detection module is used to input the city wall color image data and the expanded city wall spectral data into the disease detection model constructed by the city wall spectral information disease detection model construction method disclosed in the application for detection, so as to obtain the city wall surface disease category distribution data.
[0021] Compared with the prior art, the present invention has the following beneficial technical effects:
[0022] (1) The method for establishing a city wall spectral information disease detection model, detection method and system of the present invention eliminate the influence of the dark current of the equipment and the shooting angle during the shooting process, ensure that the spectral data finally obtained is closer to the spectral reflectance characteristics of the real object, supplement the spectral information missing in the original data, and improve the accuracy of city wall disease detection.
[0023] (2) The wall spectral information disease detection model of the present invention combines spatial information to construct more refined and specific spatial-spectral features, realizes the expansion of spectral information from sparse channels to multi-channels, and supplements the spectral information missing in the original data.
[0024] (3) The wall spectral information disease detection model of the present invention uses a feature fusion network to solve the problem of low classification accuracy caused by insufficient features under a single feature, and achieves accurate pixel-by-pixel prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a schematic diagram of the structure of the 2D module in step 2;
[0026] Figure 2 It is a schematic diagram of the structure of the 3D spatial spectrum expansion module in step 2;
[0027] Figure 3 This is the basic process of convolution in step three;
[0028] Figure 4 It is a structural diagram of the feature fusion network;
[0029] Figure 5 It is a visualization of the wall disease detection results under different algorithms in the comparative example; wherein (a) image is a color image of the target area; (b) is the disease detection accuracy under the LRDAN spectral information extension algorithm; (c) is the disease detection accuracy under the CTJN spectral information extension algorithm; (d) is the disease detection accuracy under the DCN spectral information extension algorithm; (e) is the disease detection accuracy of the method of the present invention;
[0030] Figure 6 Multi-spectral imaging data of the city wall; (a) is the sparse channel spectral data obtained by the mounted equipment; (b) is the color image data; (c) is the multi-channel spectral data obtained by the mounted equipment;
[0031] Figure 7 These are RGB images of real wall surface damage; (a) shows biological damage, (b) shows strong salt precipitation damage, (c) shows spalling damage, and (d) shows reference masonry.
[0032] Figure 8 It is the visualization of the spectral information expansion results of different algorithms. DETAILED DESCRIPTION
[0033] Specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent changes made on the basis of the technical solution of this application fall within the protection scope of the present invention.
[0034] Example 1
[0035] This embodiment discloses a method for constructing a city wall spectral information disease detection model, which specifically includes the following steps:
[0036] Step 1: obtaining a city wall spectral dataset and a city wall color image dataset, and preprocessing the city wall spectral dataset; the city wall spectral dataset includes spectral data of multiple regions of the city wall, and the city wall color image dataset includes color image data of multiple regions of the city wall;
[0037] The preprocessing of this embodiment is to perform dark current correction and registration on the city wall spectrum data set, and use the dark current calculation formula and the method of solving the transformation matrix to eliminate the influence of the device dark current and the shooting angle during the shooting process.
[0038] The preprocessing of this embodiment is as follows: in order to eliminate the interference of uneven light intensity and dark current on the spectral information, dark current and standard white board information are obtained before collecting the spectral information of the city wall, and the spectral data is corrected according to formula (1).
[0039]
[0040] In the formula, R0 is the collected original spectral data, R d is the dark current response obtained after covering the lens, R w is the calibration data obtained by shooting a standard white board, and R is the data after black and white correction. After data correction, the noise influence caused by dark current and uneven light intensity in the multi-spectral imaging system is eliminated, ensuring that the acquired spectral data is the spectral reflectance characteristics of the real object for subsequent experiments.
[0041] In the process of spectral data collection, due to slight deviations in camera position, shooting angle, and lens arrangement position, the collected spectral data needs to be registered, that is, the images of the same pixel in different bands are transformed from different coordinate systems to the same coordinate system, so that each image maintains spatial consistency with the reference image. The relevant formula is as follows:
[0042]
[0043] Where x, y are the coordinates in the original spectral image, is the corresponding coordinate in the RGB image after transformation, is the normalization factor. The rest are parameters such as rotation, scaling, and perspective in the transformation. This process solves the distortion and offset problems between multi-channel images by solving the optimal transformation matrix, so that the display of each pixel point of the multispectral imaging data of the city wall in each band maintains spatial consistency, providing conditions for the subsequent processing of the city wall spectral data.
[0044] Step 2: Input the preprocessed city wall spectral data set into the spectral information expansion network to expand and enhance the spectral dimension, and obtain the expanded city wall spectral data set; the spectral information expansion network of this embodiment includes a 2D module and a 3D spatial spectral expansion module connected in sequence. The 2D module obtains shallow spatial features, and the 3D spatial spectral expansion module extracts spatial spectral joint features of different depths in combination with the residual block of channel attention, and splices them in the feature dimension, and gradually increases the spectral information dimension through convolution and deconvolution.
[0045] Among them, the 2D module is used to extract the shallow spatial features of the city wall spectral data and output the feature map. The 2D module disclosed in this embodiment includes a 2D preliminary convolution layer, a first attention unit and a first feature fusion module connected in sequence; the first attention unit includes three first attention modules connected in parallel, each of which includes a first convolution layer (2D Conv), a first activation layer (PRelu), a second convolution layer (2D Conv), a second activation layer (PRelu) residual block and a first channel attention module connected in sequence, and the first convolution layer input end is residually connected to the output end of the second activation layer residual block. The first channel attention module (Channel Attention Mechanism) first performs global maximum pooling and global average pooling on the feature map, and then sends the two feature maps obtained by pooling into a shared multi-layer perceptron (MLP) for learning, adds the output results of the perceptron and processes them through the Sigmoid activation function mapping to obtain the final feature map. The 2D module is designed with three layers of such structures arranged in sequence. The output of each layer is fused as a fusion feature.
[0046] The 3D spatial spectrum expansion module is used to restore the spatial resolution of the feature map, supplement the missing spectral information, and output the expanded wall spectrum data. The 3D spatial spectrum expansion module disclosed in this embodiment includes a dimensionality increase layer (unsqueeze), a 3D preliminary convolution layer, a second attention unit, and a second feature fusion module connected in sequence; the second attention unit includes three second attention modules connected in parallel, each of which includes a first 3D convolution, a first PRelu activation layer, a second 3D convolution layer, a second PRelu activation layer, and a second channel attention module connected in sequence. (Channel Attention Mechanism , multi-layer perceptron, weighted layer and Sigmoid activation layer, the first 3D convolution input is residually connected to the output of the second PRelu activation layer.
[0047] The 3D spatial spectrum expansion module of this embodiment obtains spatial-spectral information from low frequency to high frequency, and realizes high-precision spectrum information expansion through multi-level feature fusion. The PReLU activation function is used in the module, which is defined as:
[0048]
[0049] The residual connection connects the input of the first 3D convolutional layer to the output of the second PRelu activation layer. Its purpose is to directly pass the spectral and spatial features extracted by the previous layer to the subsequent layers, so that the network can better integrate information in the spatial and spectral dimensions. The process of multi-level feature extraction of feature maps by the residual block is:
[0050] I res1 =PReLU(Conv2D(I up ,W res1 ,b res1 )) (4)
[0051] I res2 =PReLU(Conv2D(I res1 ,W res2 ,b res2 )) (5)
[0052] I res =I up +I res2 (6)
[0053] In the formula, the residual connection directly adds the input to the output of the subsequent layer, so that the spectral and spatial features extracted by the previous layer can be directly passed to the subsequent layer, strengthening the capture of spectral dimension features. After the residual channel attention module, the network realizes the multi-level processing and construction of spatial-spectral features, and finally uses 3D Deconvolution (3D tranConv) Restore the spatial resolution of the feature map and adjust the number of channels of the feature map. The formula is as follows:
[0054]
[0055] In the formula, W is the 3D deconvolution kernel, c represents the number of input channels, and c′ is the adjusted number of output channels. This process restores the spatial resolution of the feature map and also realizes the expansion of spectral information from sparse channels to multi-channels, supplementing the missing spectral information in the original data.
[0056] Step 3: construct a disease detection model;
[0057] The disease detection model of this embodiment includes an input layer, a convolutional layer and a feature fusion network connected in sequence;
[0058] The convolution layer includes parallel 2D convolution and 3D convolution. The 2D convolution is used to extract the color image features of the city wall color image data, and the 3D convolution is used to extract the spectral image features of the city wall spectral data. The feature fusion network includes a fusion module and a classifier connected in sequence. The fusion module is used to concatenate all color image features and all spectral image features to obtain a joint feature representation. The classifier is used to map the joint feature representation to an output dimension with the same number of classification task categories through linear transformation, and convert it into a probability distribution, and output the city wall surface disease distribution data. The classifier includes a pooling layer, a fully connected layer, and a Softmax connected in sequence.
[0059] The input layer of this embodiment receives 16 channels of multispectral imaging data of the city wall and performs a convolution operation. The convolution process is as follows: Figure 3 The calculation method of the output image size N is shown in formula (8):
[0060]
[0061] In the formula, M represents the input image size, k represents the convolution kernel size, and s represents the step size.
[0062] 2D convolution is used to extract color image features, and 3D convolution is used to extract spectral image features. After completing the extraction of spatial-spectral features and texture-color features, two different feature representations are obtained after passing through the neural network layer. Then, a feature fusion network is constructed, such as Figure 4 .
[0063] The feature fusion network first uses a pooling layer to align the spatial dimensions of the spectral features with the color image features, and then uses a convolutional layer to align the number of channels. After the alignment is completed, the RGB and multi-spectral feature maps are spliced along the channel dimension to form a joint feature representation. After high-level feature fusion learning, it is input into the classifier. The classifier contains a global pooling layer, whose function is to compress the feature map into a one-dimensional vector, reducing the number of parameters for subsequent fully connected layer operations. The fully connected layer maps the input to the same output dimension as the number of categories in the classification task through a linear transformation. Finally, the classifier uses Softmax to convert the output of the fully connected layer into a probability distribution, representing the probability value of each category, to achieve pixel-by-pixel prediction.
[0064] Step 4: using the expanded city wall spectral dataset obtained in step 2 and the city wall color image dataset obtained in step 1 as input, and using the city wall surface disease category distribution data of each region as output, the disease detection model obtained in step 3 is trained to obtain a trained disease detection model;
[0065] Among them, the categories of surface diseases of the city walls include biological diseases, severe salt precipitation diseases and peeling diseases.
[0066] Example 2
[0067] This embodiment discloses a city wall spectral information expansion disease detection method, including the following steps:
[0068] S1, collecting the city wall spectrum data and the city wall color image data, and preprocessing the city wall spectrum data;
[0069] S2, inputting the preprocessed city wall spectrum data into the spectrum information expansion network to expand and enhance the spectrum dimension, and obtaining the expanded city wall spectrum data;
[0070] S3, inputting the city wall color image data and the expanded city wall spectrum data into the disease detection model constructed by the city wall spectrum information disease detection model construction method in Example 1 for detection, and obtaining the city wall surface disease category distribution data.
[0071] Example 3
[0072] This embodiment discloses a city wall surface disease detection system, which is based on the city wall spectral information disease detection model construction method of embodiment 1 and the city wall spectral information expansion disease detection method of embodiment 2, including an acquisition module, a spectral information expansion module and a disease detection module;
[0073] The acquisition module is used to obtain the spectral data and color image data of the city wall, and to preprocess the spectral data of the city wall; the spectral information expansion module is used to expand and enhance the spectral dimension of the preprocessed spectral data of the city wall to obtain the expanded spectral data of the city wall; the disease detection module is used to input the color image data of the city wall and the expanded spectral data of the city wall for detection to obtain the distribution data of the category of diseases on the surface of the city wall.
[0074] Comparative Example
[0075] In order to verify the effect of the spectral information expansion disease detection method applied to city walls proposed in the invention in practical applications, an ancient city wall in Xi'an was selected for disease detection and classification, and the algorithm in the present invention was compared with the LRDAN algorithm, CTJN algorithm and DCN algorithm for the classification results of city wall disease detection, and the classification of the above algorithms was evaluated using the Kappa coefficient and overall accuracy. The Kappa coefficient is an accuracy evaluation index based on the confusion matrix, with a value between -1 and 1. Overall accuracy indicates the proportion of correctly classified samples to the total number of samples. The closer it is to 100%, the better the classification effect. The relevant formula can be expressed as:
[0076]
[0077] In the formula, O A Indicates the overall accuracy, T P Indicates that both the true value and the predicted value are positive samples, TN Indicates that both the true value and the predicted value are negative samples, F N Indicates that the true value is a positive sample and the predicted value is a negative sample, F P Indicates that the true value is a negative sample and the predicted value is a positive sample, K appa represents the Kappa coefficient, p o Represents the ratio of the number of correctly classified samples in each category to the total number of samples, p e represents accidental consistency, and N is the total number of samples.
[0078] In order to verify the influence of the residual channel block of the 3D spectral information extension module in the present invention on the experimental results, the original 5-channel city wall multispectral imaging data was taken as input, and experiments were carried out using the spectral information extension network constructed by ordinary convolutional layers and the extension network composed of residual channel blocks. Table 1 shows the reconstruction effect statistics of the two spectral information extension networks. RMSE, MSSIM, UIQI and SAM indices were used to evaluate and compare the performance of the extended images in terms of reconstruction accuracy, structural fidelity, overall image quality and spectral consistency.
[0079] Table 1 Comparison of spectral information expansion effects before and after using residual channel blocks
[0080]
[0081] As shown in Table 1, the results show that the extended network composed of residual channel blocks has the best effect on spectral information expansion. Compared with the spectral information expansion method using traditional convolutional layers, the mean square error and spectral angle mapping are reduced by 38.42% and 67.61%, respectively, and the average structural similarity and general image quality index are increased by 26.77% and 30.51%, respectively. The experimental results show that the introduction of residual channel blocks improves the accuracy of information expansion.
[0082] Table 2 shows the statistical results of the detection results of the four algorithms in the city wall disease detection task after spectral information expansion.
[0083] Table 2 Comparison of wall surface disease classification accuracy under different spectral information expansion algorithms
[0084]
[0085]
[0086] From the statistical table, we can see that the overall classification accuracy of the algorithm in this paper is 94.88%, and the kappa coefficient is 0.9212. The classification accuracy of common diseases is higher than that of other algorithms.
[0087] In the multispectral data of the city wall, the total number of pixels in the area to be detected is 1,457,595. Based on the reflection curve characteristics of the spectral information of the city wall surface, four algorithms are used to predict the classification of city wall diseases, and the results are visualized and compared with the actual distribution of city wall diseases. The results are as follows: Figure 5 By comparing with the pseudo-color image of the target area on the wall surface, it can be seen that the visualization effect of the algorithm in the present invention on the classification of wall surface diseases is better than that of other algorithms.
Claims
1. A method for constructing a city wall spectral information disease detection model, characterized in that: The specific steps include: Step 1: Obtain a city wall spectral dataset and a city wall color image dataset, and preprocess the city wall spectral dataset; The city wall spectrum data set includes spectrum data of city walls in multiple regions, and the city wall color image data set includes color image data of city walls in multiple regions; Step 2: Input the preprocessed city wall spectral dataset into the spectral information expansion network to expand and enhance the spectral dimension, and obtain the expanded city wall spectral dataset; Step 3: construct a disease detection model; Step 4: using the expanded city wall spectral dataset obtained in step 2 and the city wall color image dataset obtained in step 1 as input, and using the city wall surface disease category distribution data of each region as output, the disease detection model obtained in step 3 is trained to obtain a trained disease detection model; Among them, the categories of surface diseases of the city walls include biological diseases, severe salt precipitation diseases and peeling diseases.
2. The method for constructing a city wall spectral information disease detection model according to claim 1, characterized in that: The disease detection model includes an input layer, a convolutional layer and a feature fusion network connected in sequence; The convolution layer includes parallel 2D convolution and 3D convolution, the 2D convolution is used to extract the color image features of the city wall color image data, and the 3D convolution is used to extract the spectral image features of the city wall spectral data; The feature fusion network includes a fusion module and a classifier connected in sequence. The fusion module is used to splice all color image features and all spectral image features to obtain a joint feature representation. The classifier is used to map the joint feature representation to an output dimension with the same number of classification task categories through linear transformation, and convert it into a probability distribution, and output the distribution data of surface diseases of the city wall.
3. The method for constructing a city wall spectral information disease detection model according to claim 2, characterized in that: The classifier includes a pooling layer, a fully connected layer and a Softmax layer which are connected in sequence.
4. The method for constructing a city wall spectral information disease detection model according to any one of claims 1 to 3, characterized in that: The preprocessing in step 1 is to perform dark current correction and registration on the city wall spectral data set.
5. The method for constructing a city wall spectral information disease detection model according to any one of claims 1 to 3, characterized in that: The spectrum information expansion network includes a 2D module and a 3D space spectrum expansion module connected in sequence; The 2D module is used to extract the shallow spatial features of the city wall spectral data and output the feature map; The 3D spatial spectrum expansion module is used to restore the spatial resolution of the feature map, supplement the missing spectral information, and output the expanded city wall spectrum data.
6. The method for constructing a city wall spectral information disease detection model according to claim 5, characterized in that: The 2D module includes a 2D preliminary convolutional layer, a first attention unit and a first feature fusion module connected in sequence; The first attention unit includes multiple first attention modules connected in parallel, each first attention module includes a first convolutional layer, a first activation layer, a second convolutional layer, a second activation layer residual block and a first channel attention module connected in sequence, and the input end of the first convolutional layer is residually connected to the output end of the second activation layer residual block.
7. The method for constructing a city wall spectral information disease detection model according to claim 5, characterized in that: The 3D spatial spectrum expansion module includes sequentially connected Dimensional layer, 3D preliminary convolutional layer, second attention unit and second feature fusion module; The second attention unit includes multiple second attention modules connected in parallel, each second attention module includes a first 3D convolution, a first PRelu activation layer, a second 3D convolution layer, a second PRelu activation layer, a second channel attention module, a multi-layer perceptron, a weighted layer and a Sigmoid activation layer connected in sequence, and the first 3D convolution input end is residually connected to the second PRelu activation layer output end.
8. A method for detecting city wall damage by expanding spectrum information, characterized in that: The following steps are involved: S1, collecting the city wall spectrum data and the city wall color image data, and preprocessing the city wall spectrum data; S2, inputting the preprocessed city wall spectrum data into the spectrum information expansion network to expand and enhance the spectrum dimension, and obtaining the expanded city wall spectrum data; S3, inputting the city wall color image data and the expanded city wall spectral data into a disease detection model constructed by any of the city wall spectral information disease detection model construction methods as claimed in claims 1-7 for detection, and obtaining the city wall surface disease category distribution data.
9. A city wall surface disease detection system, characterized in that: Based on the city wall spectral information disease detection model construction method described in any one of claims 1 to 7 and the city wall spectral information expansion disease detection method described in claim 8, it includes an acquisition module, a spectral information expansion module and a disease detection module; The acquisition module is used to obtain the city wall spectrum data and the city wall color image data, and pre-process the city wall spectrum data; The spectral information expansion module is used to expand and enhance the spectral dimension of the preprocessed city wall spectral data to obtain expanded city wall spectral data; The disease detection module is used to input the city wall color image data and the expanded city wall spectral data into the disease detection model constructed by any city wall spectral information disease detection model construction method according to claims 1-7 for detection, so as to obtain the city wall surface disease category distribution data.