House quality evaluation method and system based on image analysis

Abstract

The invention discloses a house quality evaluation method and system based on image analysis, and relates to the technical field of image analysis, and the method comprises the steps: carrying out the preprocessing of an image set containing all defects of a house; and performing analysis processing on the house defect image training set and the convolutional neural network model to obtain a house defect image correction training set. According to the method, feature recognition is performed on the house defect image correction training set through the convolutional neural network model, the to-be-verified house defect type is determined, finally, the to-be-verified house defect type is verified through the house defect image correction training set, and the precision of the convolutional neural network model is determined. If the precision of the convolutional neural network model meets the standard, putting into use, performing defect analysis on the to-be-analyzed house image through the convolutional neural network model, determining the number of house defects, then judging the number of house defects, and determining the quality of the to-be-analyzed house, so as to avoid omission of smaller defects in the house through visual inspection; and the quality of the house can be accurately judged.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and specifically relates to a method and system for evaluating the quality of a house based on image analysis. Background Art

[0002] House defects usually refer to quality problems existing during the construction process or at the time of delivery of a house. These problems may involve multiple aspects such as the structure, facilities, and decoration of the house, and may seriously affect the safety and usability of the house when severe.

[0003] Large defects of a house can be observed by the naked eye. When there are some minor defects in the house, the naked eye cannot distinguish them. If the observation continues with the naked eye, there may be a situation of missed judgment. If there is a situation of missed judgment, the quality of the house cannot be accurately judged. Summary of the Invention

[0004] To solve the above technical problems, a method and system for evaluating the quality of a house based on image analysis are provided. The present technical solution solves the problem that large defects of a house can be observed by the naked eye, but when there are some minor defects in the house, the naked eye cannot distinguish them. If the observation continues with the naked eye, there may be a situation of missed judgment, and if there is a situation of missed judgment, the quality of the house cannot be accurately judged as described in the above background art.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A method for evaluating the quality of a house based on image analysis, comprising:

[0007] Obtain an image set containing all defects of the house, and based on a quality evaluation terminal, preprocess the image set containing all defects of the house to obtain a house defect image training set;

[0008] Based on the quality evaluation terminal, analyze and process the house defect image training set and a convolutional neural network model to obtain a corrected house defect image training set;

[0009] Based on the quality evaluation terminal, analyze and process the corrected house defect image training set to determine the filter parameters of the convolutional layer;

[0010] Based on the quality evaluation terminal, input the corrected house defect image training set into the convolutional neural network model for analysis and processing to obtain the types of house defects to be verified;

[0011] The quality evaluation terminal verifies the types of house defects to be verified through the corrected house defect image training set to determine the accuracy of the convolutional neural network model;

[0012] Collect images of the house to be analyzed through an image capture device to obtain images of the house to be analyzed;

[0013] Based on a convolutional neural network model, analyze and process the images of the house to be analyzed to determine the quality of the house to be analyzed.

[0014] Preferably, the steps of obtaining an image set containing all defects of the house, preprocessing the image set containing all defects of the house based on a quality evaluation terminal, and obtaining a training set of house defect images are specifically as follows:

[0015] Based on a quality evaluation terminal, perform data extraction processing on the database system to obtain an image set containing all defects of the house;

[0016] Based on the discrete Fourier transform algorithm, convert the images in the image set containing all defects of the house from the spatial domain to the frequency domain to obtain the frequency information of the images in the image set containing all defects of the house;

[0017] Based on a quality evaluation terminal, analyze and process the frequency information of the images in the image set containing all defects of the house to obtain a preprocessed image set of house defects;

[0018] Based on a quality evaluation terminal, perform rotation processing and color transformation processing on the images in the preprocessed image set of house defects to obtain a training set of house defect images;

[0019] Among them, the specific calculation formula for obtaining the frequency information of the images in the image set containing all defects of the house is:

[0020]

[0021] In the formula, F(u, v) is the frequency information of the images in the image set containing all defects of the house; u is the row coordinate of the frequency value of the images in the image set containing all defects of the house; v is the column coordinate of the frequency value of the images in the image set containing all defects of the house; f(x, y) is the pixel value of the images in the image set containing all defects of the house; M is the horizontal size of the images in the image set containing all defects of the house; N is the vertical size of the images in the image set containing all defects of the house; x is the row coordinate of the pixel value of the images in the image set containing all defects of the house; y is the column coordinate of the pixel value of the images in the image set containing all defects of the house.

[0022] Preferably, the steps of analyzing and processing the frequency information of the images in the image set containing all defects of the house based on a quality evaluation terminal to obtain a preprocessed image set of house defects are specifically as follows:

[0023] Based on a band-pass filter, denoise the frequency information of the images in the image set containing all the defects of the house to obtain the frequency information of the house defect images after the first preprocessing;

[0024] Based on a quality assessment terminal, perform spectral enhancement processing on the frequency information of the house defect images after the first preprocessing to obtain the frequency information of the house defect images after the second preprocessing;

[0025] Based on the inverse Fourier transform algorithm, convert the frequency information of the house defect images after the second preprocessing from the frequency domain to the spatial domain to obtain the preprocessed house defect image set.

[0026] Preferably, the analysis and processing of the house defect image training set and the convolutional neural network model based on the quality assessment terminal to obtain the corrected training set of the house defect images specifically includes the following steps:

[0027] Based on the quality assessment terminal, perform size analysis processing on the images in the house defect image training set to determine the image size information in the house defect image training set;

[0028] Based on the quality assessment terminal, perform size analysis processing on the input layer of the convolutional neural network model to determine the size information of the input data;

[0029] Based on the quality assessment terminal, perform calculation processing on the image size information in the house defect image training set and the size information of the input data to determine the image scaling ratio;

[0030] The quality assessment terminal corrects the image size information in the house defect image training set according to the image scaling ratio to obtain the corrected training set of the house defect images;

[0031] Among them, the specific calculation formula for determining the image scaling ratio is:

[0032]

[0033] In the formula, I d is the image scaling ratio; I1 is the image size information in the house defect image training set; I2 is the size information of the input data.

[0034] Preferably, the analysis and processing of the corrected training set of the house defect images based on the quality assessment terminal to determine the filter parameters of the convolutional layer specifically includes the following steps:

[0035] Based on the quality assessment terminal, perform image channel analysis on the images in the corrected training set of the house defect images to determine the number of image channels;

[0036] The quality assessment terminal designs the filters of the convolutional layer according to the number of image channels to determine the filter size information of the convolutional layer;

[0037] Based on the quality assessment terminal, analyze and process the filter size information of the convolutional layer and the size information of the input data to determine the filter stride information of the convolutional layer.

[0038] Preferably, the above-mentioned quality assessment terminal inputs the corrected training set of house defect images into the convolutional neural network model for analysis and processing. The specific steps for obtaining the house defect types to be verified are as follows:

[0039] Based on the filters of the convolutional layer, perform convolutional calculation processing on each region of the images in the corrected training set of house defect images to obtain several groups of eigenvalue of house defect images;

[0040] Based on the convolutional neural network model, perform image construction on the eigenvalues of several groups of house defect images to obtain an initial feature map;

[0041] Based on the ReLU activation function, perform activation processing on the elements in the initial feature map to obtain an activated feature map;

[0042] Based on the convolutional neural network model, analyze and process the activated feature map to obtain the house defect types to be verified;

[0043] Among them, the specific calculation formula for obtaining the activated feature map is:

[0044] h(x) = max(0, x);

[0045] In the formula, h(x) is the activated feature map; x is the element in the initial feature map; max() is the maximum value function.

[0046] Preferably, the above-mentioned analysis and processing of the activated feature map based on the convolutional neural network model to obtain the house defect types to be verified specifically include the following steps:

[0047] Based on the convolutional neural network model, perform size analysis on the activated feature map to determine the size information of the activated feature map;

[0048] Based on the convolutional neural network model, analyze and process the size information of the activated feature map to determine the pooling stride of the pooling layer;

[0049] Based on the max pooling algorithm and the pooling stride, perform feature screening on the activated feature map to obtain a pooled feature map;

[0050] Based on the convolutional neural network model, perform dimensional flattening processing on the pooled feature map to obtain one-dimensional feature data;

[0051] The fully connected layer of the convolutional neural network model performs computational processing on one-dimensional feature data through matrix multiplication to obtain a one-dimensional feature vector;

[0052] Based on the Softmax function, classify the one-dimensional feature vector to determine the type of housing defect to be verified.

[0053] Preferably, the quality evaluation terminal verifies the type of housing defect to be verified through the housing defect image correction training set, and determining the accuracy of the convolutional neural network model specifically includes the following steps:

[0054] The quality evaluation terminal performs matching processing on the housing defect image correction training set through the type of housing defect to be verified;

[0055] If there is an image associated with the type of housing defect to be verified in the housing defect image correction training set, the convolutional neural network model can accurately identify the type of defect in the housing image;

[0056] If there is no image associated with the type of housing defect to be verified in the housing defect image correction training set, change the parameters inside the convolutional neural network model and retrain the convolutional neural network model, and obtain the type of housing defect to be verified again for verification.

[0057] Preferably, the convolutional neural network model is used to analyze and process the housing image to be analyzed, and determining the quality of the housing to be analyzed specifically includes the following steps:

[0058] Input the housing image to be analyzed into the convolutional neural network model to determine the number of housing defects;

[0059] Based on the quality evaluation terminal, judge and process the number of housing defects and the set defect number threshold;

[0060] If the number of housing defects is less than the set first defect number threshold, the quality of the housing to be analyzed is excellent;

[0061] If the number of housing defects is greater than or equal to the set first defect number threshold and less than the set second defect number threshold, the quality of the housing to be analyzed is good;

[0062] If the number of housing defects is greater than or equal to the set second defect number threshold, the quality of the housing to be analyzed is poor.

[0063] Furthermore, a housing quality evaluation system based on image analysis is proposed, which is used to implement a housing quality evaluation method based on image analysis as described above, including:

[0064] Quality assessment terminal. The quality assessment terminal designs a convolutional neural network model by analyzing an image set containing all defects of a house. The quality assessment terminal is used to control data transmission and information interaction among various modules;

[0065] Image preprocessing module. The image preprocessing module denoises and enhances an image set containing all defects of a house to obtain a preprocessed image set of house defects;

[0066] Image size adjustment module. The image size adjustment module is used to adjust the size of images in the preprocessed image set of house defects to determine a corrected training set of house defect images;

[0067] Convolutional neural network model. The convolutional neural network model is used to perform feature analysis on images in the corrected training set of house defect images to obtain types of house defects to be verified;

[0068] House defect verification module. The house defect verification module verifies the types of house defects to be verified according to the corrected training set of house defect images to determine the accuracy of the convolutional neural network model;

[0069] Image capture device. The image capture device is used to capture images of a house to be analyzed to obtain images of the house to be analyzed;

[0070] Quality assessment module. The quality assessment module judges and processes the number of house defects and a set defect number threshold to determine the quality of the house to be analyzed.

[0071] Compared with the prior art, the present invention provides a method and system for house quality assessment based on image analysis, having the following beneficial effects:

[0072] First, the present invention preprocesses an image set containing all defects of a house, removes noise in the images, and at the same time enhances the features of the images, improving the recognition of house defect features. Then, it rotates and performs color transformation on the preprocessed image set of house defects, expanding the training set of house defect images. Next, it performs feature recognition on the corrected training set of house defect images through the convolutional neural network model to determine the types of house defects to be verified. Finally, it verifies the types of house defects to be verified through the corrected training set of house defect images to determine the accuracy of the convolutional neural network model. If the accuracy of the convolutional neural network model meets the standard, it is put into use. It analyzes the defects of the images of the house to be analyzed through the convolutional neural network model to determine the number of house defects, and then judges the number of house defects to determine the quality of the house to be analyzed, avoiding missing smaller defects in the house by visual observation and being able to accurately judge the quality of the house. Description of the Drawings

[0073] Figure 1Schematic flowchart of steps S100 - S700 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0074] Figure 2 Schematic flowchart of steps S101 - S104 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0075] Figure 3 Schematic flowchart of steps S1031 - S1033 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0076] Figure 4 Schematic flowchart of steps S201 - S204 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0077] Figure 5 Schematic flowchart of steps S301 - S303 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0078] Figure 6 Schematic flowchart of steps S401 - S404 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0079] Figure 7 Schematic flowchart of steps S4041 - S4046 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0080] Figure 8 Schematic flowchart of steps S501 - S503 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0081] Figure 9 Schematic flowchart of steps S501 - S503 in a method for evaluating housing quality based on image analysis proposed by the present invention;

[0082] Figure 10 Block diagram of the structure of a housing quality evaluation system based on image analysis proposed by the present invention. Detailed implementation manners

[0083] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations.

[0084] Referring to Figure 1 as shown, a method for evaluating housing quality based on image analysis includes:

[0085] S100. Obtain an image set containing all defects of a house, and based on a quality assessment terminal, preprocess the image set containing all defects of the house to obtain a training set of house defect images;

[0086] S200. Based on the quality assessment terminal, analyze and process the training set of house defect images and a convolutional neural network model to obtain a corrected training set of house defect images;

[0087] S300. Based on the quality assessment terminal, analyze and process the corrected training set of house defect images to determine the filter parameters of the convolutional layer;

[0088] S400. Based on the quality assessment terminal, input the corrected training set of house defect images into the convolutional neural network model for analysis and processing to obtain the house defect types to be verified;

[0089] S500. The quality assessment terminal verifies the house defect types to be verified through the corrected training set of house defect images to determine the accuracy of the convolutional neural network model;

[0090] S600. Collect images of the house to be analyzed through an image capture device to obtain images of the house to be analyzed;

[0091] S700. Based on the convolutional neural network model, analyze and process the images of the house to be analyzed to determine the quality of the house to be analyzed;

[0092] Those skilled in the art can understand that large defects of a house can be identified and judged by the naked eye. When there are some subtle defects in the house, the naked eye cannot accurately judge those subtle defects. If the observation continues with the naked eye, there may be a situation of missed judgment, and thus the quality of the house cannot be accurately judged. Therefore, train the convolutional neural network model through an image set containing all defects of the house, then analyze the defects of the images of the house to be analyzed through the trained convolutional neural network model to obtain the number of house defects. Finally, judge the number of house defects to accurately obtain the quality of the house.

[0093] Refer to Figure 2 As shown, obtaining an image set containing all defects of a house, and based on a quality assessment terminal, preprocessing the image set containing all defects of the house to obtain a training set of house defect images specifically includes the following steps:

[0094] S101. Based on the quality assessment terminal, perform data extraction processing on the database system to obtain an image set containing all defects of the house;

[0095] S102. Based on the discrete Fourier transform algorithm, convert the images in the image set containing all defects of the house from the spatial domain to the frequency domain to obtain the frequency information of the images in the image set containing all defects of the house;

[0096] S103. Based on the quality assessment terminal, analyze and process the frequency information of the images in the image set containing all housing defects to obtain a preprocessed housing defect image set;

[0097] S104. Based on the quality assessment terminal, perform rotation processing and color transformation processing on the images in the preprocessed housing defect image set to obtain a housing defect image training set;

[0098] It can be understood that the images in the preprocessed housing defect image set are rotated at different angles, such as 30° rotation, 45° rotation, etc. The rotated images are only different in angle from the original images. The background colors of the images in the preprocessed housing defect image set can also be changed to obtain images with different background colors. The above operations are to expand the housing defect image training set and improve the feature recognition accuracy of the convolutional neural network model;

[0099] Among them, the specific calculation formula for obtaining the frequency information of the images in the image set containing all housing defects is:

[0100]

[0101] In the formula, F(u, v) is the frequency information of the images in the image set containing all housing defects; u is the row coordinate of the frequency value of the images in the image set containing all housing defects; v is the column coordinate of the frequency value of the images in the image set containing all housing defects; f(x, y) is the pixel value of the images in the image set containing all housing defects; M is the horizontal size of the images in the image set containing all housing defects; N is the vertical size of the images in the image set containing all housing defects; x is the row coordinate of the pixel value of the images in the image set containing all housing defects; y is the column coordinate of the pixel value of the images in the image set containing all housing defects;

[0102] In this embodiment, the preprocessing of the image set containing all housing defects is to remove the interference factors in the image and improve the recognition accuracy of the convolutional neural network model. In addition, the rotation and color conversion of the preprocessed images are to expand the training image set, so that the convolutional neural network model can recognize the same images in different environments, indirectly improving the accuracy of the convolutional neural network model and avoiding multiple trainings of the convolutional neural network model. In addition, the noises of the images in the image set containing all housing defects are all of fixed frequencies. Therefore, a band-pass filter is designed to denoise them.

[0103] Refer to Figure 3 As shown, based on the quality assessment terminal, the analysis and processing of the frequency information of the images in the image set containing all housing defects to obtain a preprocessed housing defect image set specifically includes the following steps:

[0104] S1031. Based on a band - pass filter, denoise the frequency information of the images in the image set containing all housing defects to obtain the frequency information of the housing defect images after the first pre - processing.

[0105] S1032. Based on a quality assessment terminal, perform spectral enhancement processing on the frequency information of the housing defect images after the first pre - processing to obtain the frequency information of the housing defect images after the second pre - processing.

[0106] S1033. Based on the inverse Fourier transform algorithm, convert the frequency information of the housing defect images after the second pre - processing from the frequency domain to the spatial domain to obtain the pre - processed housing defect image set.

[0107] In this embodiment, the image set containing all housing defects may contain noise, and the noise in the images may affect the accuracy of the convolutional neural network model. Therefore, denoise the image set containing all housing defects. At the same time, in order to make the convolutional neural network model more accurate in recognizing the housing defect features in the images, enhance the image features in the image set used for training.

[0108] Refer to Figure 4 As shown, based on a quality assessment terminal, analyze and process the housing defect image training set and the convolutional neural network model to obtain the corrected housing defect image training set, which specifically includes the following steps:

[0109] S201. Based on a quality assessment terminal, perform size analysis processing on the images in the housing defect image training set to determine the image size information in the housing defect image training set.

[0110] S202. Based on a quality assessment terminal, perform size analysis processing on the input layer of the convolutional neural network model to determine the size information of the input data.

[0111] S203. Based on a quality assessment terminal, perform calculation processing on the image size information in the housing defect image training set and the size information of the input data to determine the image scaling ratio.

[0112] S204. The quality assessment terminal corrects the image size information in the housing defect image training set according to the image scaling ratio to obtain the corrected housing defect image training set.

[0113] Among them, the specific calculation formula for determining the image scaling ratio is:

[0114]

[0115] In the formula, I d is the image scaling ratio; I1 is the image size information in the housing defect image training set; I2 is the size information of the input data.

[0116] In this embodiment, the input layer of the convolutional neural network model can only receive images of a fixed size. Therefore, analyze the input layer of the convolutional neural network model to determine the size information of the input data. Then, calculate and process the size information of the images in the house defect image training set and the size information of the input data to determine the image scaling ratio. Resize the images in the house defect image training set according to the image scaling ratio to obtain house defect images adapted to the input layer of the convolutional neural network model.

[0117] Refer to Figure 5 As shown, based on the quality assessment terminal, analyzing and processing the corrected training set of house defect images to determine the filter parameters of the convolutional layer specifically includes the following steps:

[0118] S301. Based on the quality assessment terminal, perform image channel analysis on the images in the corrected training set of house defect images to determine the number of image channels;

[0119] S302. The quality assessment terminal designs the filters of the convolutional layer according to the number of image channels to determine the size information of the filters of the convolutional layer;

[0120] S303. Based on the quality assessment terminal, analyze and process the size information of the filters of the convolutional layer and the size information of the input data to determine the stride information of the filters of the convolutional layer;

[0121] In this embodiment, the size and depth of the filter are related to the channels of the image. For example, an RGB image has three channels: red, green, and blue. Therefore, the size information of the filter should be designed as 3*3*3. Therefore, it is necessary to analyze the number of image channels to determine the size information of the filter, and the stride information of the filter determines the size of the initial feature map. In order to keep the size of the feature map unchanged, either pad the feature map or change the stride of the filter. However, the complexity of padding the feature map is greater than the design of the stride of the filter. Therefore, ensure the size of the feature map remains unchanged by designing the stride of the filter.

[0122] Refer to Figure 6 As shown, based on the quality assessment terminal, input the corrected training set of house defect images into the convolutional neural network model for analysis and processing to obtain the house defect types to be verified, specifically including the following steps:

[0123] S401. Based on the filters of the convolutional layer, perform convolutional calculation processing on each region of the images in the corrected training set of house defect images to obtain several sets of eigenvalue of house defect images;

[0124] S402. Based on the convolutional neural network model, perform image construction on the eigenvalues of several sets of house defect images to obtain the initial feature map;

[0125] S403. Activate the elements in the initial feature map based on the ReLU activation function to obtain an activated feature map.

[0126] S404. Analyze and process the activated feature map based on the convolutional neural network model to obtain the type of housing defect to be verified.

[0127] Among them, the specific calculation formula for obtaining the activated feature map is:

[0128] h(x) = max(0, x);

[0129] In the formula, h(x) is the activated feature map; x is the element in the initial feature map; max() is the maximum value function.

[0130] In this embodiment, in order to reduce the subsequent calculation amount, the initial feature map is activated by the ReLU activation function to improve the sparsity of the initial feature map, changing some elements in the initial feature map to zero, reducing the subsequent calculation amount of image feature recognition. Because the pixel values of the image are between [-1, 1], when the image is relatively smooth, that is, there are no defects in the image, the pixel values of the image are small. Therefore, in order to reduce the calculation amount, the pixel values in the relatively smooth area are filtered out.

[0131] Refer to Figure 7 As shown, based on the convolutional neural network model, analyzing and processing the activated feature map to obtain the type of housing defect to be verified specifically includes the following steps:

[0132] S4041. Based on the convolutional neural network model, perform size analysis on the activated feature map to determine the size information of the activated feature map.

[0133] S4042. Based on the convolutional neural network model, analyze and process the size information of the activated feature map to determine the pooling step of the pooling layer.

[0134] S4043. Based on the max-pooling algorithm and the pooling step, perform feature screening on the activated feature map to obtain a pooled feature map.

[0135] S4044. Based on the convolutional neural network model, perform dimensional flattening processing on the pooled feature map to obtain one-dimensional feature data.

[0136] S4045. Based on the fully connected layer of the convolutional neural network model, perform calculation processing on the one-dimensional feature data through matrix multiplication to obtain a one-dimensional feature vector.

[0137] S4046. Based on the Softmax function, perform classification processing on the one-dimensional feature vector to determine the type of housing defect to be verified.

[0138] In this embodiment, the more obvious the features in the image are, the larger the pixel values of the image. Therefore, the activation feature map is pooled by the pooling layer, and the maximum values within a certain range are selected through the maximum pooling algorithm, which are the features of the image. However, when some features of the image change violently within a certain range, there will be more relatively large feature values. To make feature recognition more accurate, by analyzing the size information of the activation feature map, the pooling step size of the pooling layer is reduced. For example, the original data matrix is Figure 1 and the pooling step size is 2*2, then the pooled feature map is and the pooled feature map represents the features in the image. In addition, since the output of the convolutional layer or the pooling layer is a two-dimensional feature map, while the fully connected layer requires a one-dimensional input, first, the two-dimensional feature map needs to be flattened, converting the multi-dimensional input data into one-dimensional data. The flattened feature map is input into the fully connected layer and converted into a one-dimensional feature vector of a fixed length through matrix multiplication. This process realizes the mapping from the high-dimensional feature space to the low-dimensional feature space while retaining the most useful information. Finally, the fully connected layer performs classification or regression operations on the one-dimensional feature vector through an activation function (such as the Softmax function). In the classification task, the Softmax function converts the feature vector into a probability distribution, representing the possibility of each housing defect; then, the possibility of each housing defect is judged, and the type of housing defect to be verified is output. As shown in

[0139] reference Figure 8 shown, the quality assessment terminal verifies the type of housing defect to be verified through the housing defect image correction training set, and determining the accuracy of the convolutional neural network model specifically includes the following steps:

[0140] S501. The quality assessment terminal performs a matching process on the housing defect image correction training set according to the type of housing defect to be verified;

[0141] S502. If there is an image associated with the type of housing defect to be verified in the housing defect image correction training set, the convolutional neural network model can accurately identify the defect type of the housing image;

[0142] S503. If there is no image associated with the type of housing defect to be verified in the housing defect image correction training set, change the parameters inside the convolutional neural network model and retrain the convolutional neural network model, and obtain the type of housing defect to be verified again for verification.

[0143] As shown in Figure 9 reference

[0144] S701. Input the image of the house to be analyzed into the convolutional neural network model to determine the number of house defects;

[0145] S702. Based on the quality assessment terminal, judge and process the number of house defects and the set defect number threshold;

[0146] S703. If the number of house defects is less than the set first defect number threshold, the quality of the house to be analyzed is excellent;

[0147] S704. If the number of house defects is greater than or equal to the set first defect number threshold and less than the set second defect number threshold, the quality of the house to be analyzed is good;

[0148] S705. If the number of house defects is greater than or equal to the set second defect number threshold, the quality of the house to be analyzed is poor;

[0149] In this embodiment, before the convolutional neural network model is put into use, it is necessary to analyze the accuracy of its judgment. If the accuracy of the convolutional neural network model is insufficient, it needs to be retrained. If the convolutional neural network model with insufficient accuracy is put into use, there may be cases of missed judgment or misjudgment of house defects. If there are cases of missed judgment or misjudgment of house defects, the quality of the house cannot be accurately judged.

[0150] Refer to Figure 10 As shown, a house quality assessment system based on image analysis is used to implement a house quality assessment method based on image analysis as described above, including:

[0151] A quality assessment terminal, which designs a convolutional neural network model by analyzing an image set containing all defects of the house, and the quality assessment terminal is used to control data transmission and information interaction between each module;

[0152] An image preprocessing module, which denoises and enhances the image set containing all defects of the house to obtain a preprocessed image set of house defects;

[0153] An image size adjustment module, which is used to adjust the size of the images in the preprocessed image set of house defects to determine a corrected training set of house defect images;

[0154] A convolutional neural network model, which is used to perform feature analysis on the images in the corrected training set of house defect images to obtain the types of house defects to be verified;

[0155] A house defect verification module, which verifies the types of house defects to be verified according to the corrected training set of house defect images to determine the accuracy of the convolutional neural network model;

[0156] An image capturing device for capturing images of a house to be analyzed and obtaining images of the house to be analyzed;

[0157] A quality assessment module for judging and processing the number of house defects and a set defect number threshold to determine the quality of the house to be analyzed;

[0158] In this embodiment, the number of house defects affects the quality of the house. Therefore, the quality of the house is determined by judging the house defect data.

[0159] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the quality of a house based on image analysis, characterized in that, Including: Obtain an image set containing all defects of the house, and based on the quality assessment terminal, preprocess the image set containing all defects of the house to obtain a training set of house defect images; Based on the quality assessment terminal, analyze and process the training set of house defect images and the convolutional neural network model to obtain a corrected training set of house defect images; Based on the quality assessment terminal, analyze and process the corrected training set of house defect images to determine the filter parameters of the convolutional layer; Based on the quality assessment terminal, input the corrected training set of house defect images into the convolutional neural network model for analysis and processing to obtain the house defect types to be verified; The quality assessment terminal verifies the house defect types to be verified through the corrected training set of house defect images to determine the accuracy of the convolutional neural network model; Collect images of the house to be analyzed through an image capture device to obtain images of the house to be analyzed; Based on the convolutional neural network model, analyze and process the images of the house to be analyzed to determine the quality of the house to be analyzed.

2. The method for evaluating the quality of a house based on image analysis according to claim 1, wherein The specific steps for obtaining an image set containing all defects of the house and preprocessing the image set containing all defects of the house based on the quality assessment terminal to obtain a training set of house defect images are as follows: Based on the quality assessment terminal, perform data extraction processing on the database system to obtain an image set containing all defects of the house; Based on the discrete Fourier transform algorithm, convert the images in the image set containing all defects of the house from the spatial domain to the frequency domain to obtain the frequency information of the images in the image set containing all defects of the house; Based on the quality assessment terminal, analyze and process the frequency information of the images in the image set containing all defects of the house to obtain a preprocessed image set of house defects; Based on the quality assessment terminal, perform rotation processing and color transformation processing on the images in the preprocessed image set of house defects to obtain a training set of house defect images; Among them, the specific calculation formula for obtaining the frequency information of the images in the image set containing all defects of the house is: In the formula, F(u, v) is the frequency information of the images in the image set containing all defects of the house; u is the row coordinate of the frequency value of the images in the image set containing all defects of the house; v is the column coordinate of the frequency value of the images in the image set containing all defects of the house; f(x, y) is the pixel value of the images in the image set containing all defects of the house; M is the horizontal size of the images in the image set containing all defects of the house; N is the vertical size of the images in the image set containing all defects of the house; x is the row coordinate of the pixel value of the images in the image set containing all defects of the house; y is the column coordinate of the pixel value of the images in the image set containing all defects of the house.

3. The method for evaluating the quality of a house based on image analysis according to claim 2, wherein The specific steps for analyzing and processing the frequency information of the images in the image set containing all defects of the house based on the quality assessment terminal to obtain a preprocessed image set of house defects are as follows: Based on a band-pass filter, perform noise reduction processing on the frequency information of the images in the image set containing all defects of the house to obtain the frequency information of the house defect images after the first preprocessing; Based on the quality assessment terminal, perform spectral enhancement processing on the frequency information of the first preprocessed house defect image to obtain the frequency information of the second preprocessed house defect image; Based on the inverse Fourier transform algorithm, convert the frequency information of the second preprocessed house defect image from the frequency domain to the spatial domain to obtain the preprocessed house defect image set.

4. A method for evaluating the quality of a house based on image analysis according to claim 1, characterized in that, The analysis and processing of the house defect image training set and the convolutional neural network model based on the quality assessment terminal to obtain the corrected house defect image training set specifically includes the following steps: Based on the quality assessment terminal, perform size analysis processing on the images in the house defect image training set to determine the image size information in the house defect image training set; Based on the quality assessment terminal, perform size analysis processing on the input layer of the convolutional neural network model to determine the size information of the input data; Based on the quality assessment terminal, perform calculation processing on the image size information in the house defect image training set and the size information of the input data to determine the image scaling ratio; The quality assessment terminal corrects the image size information in the house defect image training set according to the image scaling ratio to obtain the corrected house defect image training set; Among them, the specific calculation formula for determining the image scaling ratio is: Where, I d is the image scaling ratio; I1 is the image size information in the training set of housing defect images; I2 is the size information of the input data.

5. A method for evaluating the quality of a house based on image analysis according to claim 1, characterized in that, The analysis and processing of the corrected house defect image training set based on the quality assessment terminal to determine the filter parameters of the convolutional layer specifically includes the following steps: Based on the quality assessment terminal, perform image channel analysis on the images in the corrected house defect image training set to determine the number of image channels; The quality assessment terminal designs the filters of the convolutional layer according to the number of image channels to determine the filter size information of the convolutional layer; Based on the quality assessment terminal, perform analysis processing on the filter size information of the convolutional layer and the size information of the input data to determine the filter stride information of the convolutional layer.

6. A method for evaluating the quality of a house based on image analysis according to claim 1, characterized in that, The input of the corrected house defect image training set into the convolutional neural network model for analysis and processing based on the quality assessment terminal to obtain the house defect type to be verified specifically includes the following steps: Based on the filters of the convolutional layer, perform convolutional calculation processing on each region of the images in the corrected house defect image training set to obtain several sets of eigenvalue of the house defect images; Based on the convolutional neural network model, perform image construction on the eigenvalues of several sets of house defect images to obtain the initial feature map; Based on the ReLU activation function, perform activation processing on the elements in the initial feature map to obtain the activation feature map; Based on the convolutional neural network model, perform analysis processing on the activation feature map to obtain the house defect type to be verified; Among them, the specific calculation formula for obtaining the activation feature map is: h(x) = max(0, x); In the formula, h(x) is the activation feature map; x is the element in the initial feature map; max() is the maximum value function.

7. The method for evaluating the quality of a house based on image analysis according to claim 6, characterized in that, The analysis and processing of the activation feature map based on the convolutional neural network model to obtain the house defect type to be verified specifically includes the following steps: Based on the convolutional neural network model, perform size analysis on the activation feature map to determine the size information of the activation feature map; Based on the convolutional neural network model, analyze and process the size information of the activation feature map to determine the pooling stride of the pooling layer; Based on the max pooling algorithm and the pooling stride, perform feature screening on the activation feature map to obtain the pooled feature map; Based on the convolutional neural network model, perform dimensional flattening processing on the pooled feature map to obtain one-dimensional feature data; Based on the fully connected layer of the convolutional neural network model, perform calculation processing on the one-dimensional feature data through matrix multiplication to obtain a one-dimensional feature vector; Based on the Softmax function, perform classification processing on the one-dimensional feature vector to determine the type of housing defect to be verified.

8. A method for evaluating the quality of a house based on image analysis according to claim 1, characterized in that The quality evaluation terminal verifies the type of housing defect to be verified through the housing defect image correction training set, and determining the accuracy of the convolutional neural network model specifically includes the following steps: The quality evaluation terminal performs matching processing on the housing defect image correction training set through the type of housing defect to be verified; If there is an image associated with the type of housing defect to be verified in the housing defect image correction training set, the convolutional neural network model can accurately identify the defect type of the housing image; If there is no image associated with the type of housing defect to be verified in the housing defect image correction training set, change the parameters inside the convolutional neural network model and retrain the convolutional neural network model, and obtain the type of housing defect to be verified again for verification.

9. The method for evaluating the quality of a house based on image analysis according to claim 1, wherein, The method for analyzing and processing the housing image to be analyzed based on the convolutional neural network model to determine the quality of the housing to be analyzed specifically includes the following steps: Input the housing image to be analyzed into the convolutional neural network model to determine the number of housing defects; Based on the quality evaluation terminal, perform judgment processing on the number of housing defects and the set defect number threshold; If the number of housing defects is less than the set first defect number threshold, the quality of the housing to be analyzed is excellent; If the number of housing defects is greater than or equal to the set first defect number threshold and less than the set second defect number threshold, the quality of the housing to be analyzed is good; If the number of housing defects is greater than or equal to the set second defect number threshold, the quality of the housing to be analyzed is poor.

10. A housing quality assessment system based on image analysis, which is used to implement a housing quality assessment method based on image analysis as described in any one of claims 1-9, characterized in that, Including: A quality evaluation terminal, which designs a convolutional neural network model by analyzing an image set containing all housing defects, and the quality evaluation terminal is used to control data transmission and information interaction between each module; An image preprocessing module, which denoises and enhances the image set containing all housing defects to obtain a preprocessed housing defect image set; An image size adjustment module, which is used to adjust the size of the images in the preprocessed housing defect image set to determine the housing defect image correction training set; A convolutional neural network model, which is used to perform feature analysis on the images in the housing defect image correction training set to obtain the type of housing defect to be verified; A housing defect verification module, which verifies the type of housing defect to be verified according to the housing defect image correction training set to determine the accuracy of the convolutional neural network model; An image capture device, which is used to capture images of the housing to be analyzed to obtain the housing image to be analyzed; A quality assessment module, which judges and processes the number of housing defects and a set defect quantity threshold to determine the quality of the house to be analyzed.

Images