Building progress evaluation method based on construction image

Abstract

The invention discloses a construction image-based building progress evaluation method, which is used in the field of building construction, and comprises the following steps of: configuring a wireless communication module of an unmanned aerial vehicle and a meteorological measurement station; acquiring meteorological parameter data and meteorological images; formulating a flight scheme under the abnormal weather according to the abnormal weather in the comprehensive prediction; dark channel defogging is utilized to improve the quality of an abnormal building construction image, and a SegNet deep network is utilized to extract key information of a building progress; comparing the key information of the building progress with the actual progress information, the planned progress information and the historical progress information of the same period in the previous time period; and visually presenting a building progress evaluation result based on the construction image through 3D modeling. According to the method, the time sequence model is established, key features are extracted, the weather condition of the next week is comprehensively predicted, weather planning is made in advance, and corresponding measures are taken to deal with abnormal weather possibly occurring.

Description

Technical Field

[0001] The present invention relates to the field of building construction, and in particular to a building progress assessment method based on construction images. Background Art

[0002] The construction industry is the cornerstone of the national economy and plays an indispensable role in economic and social progress. It is generally believed that accurate supervision of construction is a key factor in ensuring the success of construction projects, and the core of such supervision is the accurate assessment of construction progress. The construction phase is a critical period in the life cycle of a construction project, and most of the time and funds are invested in this phase. Therefore, progress assessment and management of the construction phase is crucial to achieving project goals.

[0003] The assessment of construction progress is an ongoing process that requires the continuous collection and updating of data from the site. This results in a large flow of information, which continues to increase as the project progresses. The dynamics and volume of this real-time information poses challenges to the storage, sharing and management of information. Traditional methods rely mainly on manual collection of field data through manual measurements and written records, and then processing this data and generating reports. This traditional data collection method can no longer meet the needs of modern construction projects, and its limitations are becoming increasingly apparent.

[0004] In the modern construction industry, drone technology has become an effective tool to assist in the assessment and monitoring of building construction progress. Drones have many advantages, especially in data collection and monitoring, but they also have some potential limitations.

[0005] First, the use of drones is significantly more convenient in construction. Drones can cover a wide range of construction areas in a relatively short period of time, capturing high-resolution image and video data. This data can be used to generate a 3D model of the construction site, helping supervisors and management teams better understand the actual progress of the project. In addition, drones can fly regularly and collect data in real time, providing accurate information about construction progress and site conditions.

[0006] However, there are also some limitations to drone technology, one of which is that it is affected by adverse weather and environmental conditions. Adverse weather conditions, such as strong winds, heavy rain or fog, may limit the flight capabilities of drones, thereby affecting the frequency and quality of data collection. This may lead to inaccuracy and discontinuity in the data, thereby reducing the ability to accurately assess construction progress.

[0007] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0008] In order to overcome the above problems, the present invention aims to propose a construction progress assessment method based on construction images, aiming to solve the problem that severe weather conditions, such as strong winds, heavy rains or fog, may limit the flight capabilities of drones, thereby affecting the frequency and quality of data collection.

[0009] To this end, the specific technical solution adopted by the present invention is as follows:

[0010] A construction progress assessment method based on construction images, the construction progress assessment method based on construction images comprises the following steps:

[0011] S1. Set up a meteorological measurement station at the construction site and equip the drone with a wireless communication module for the meteorological measurement station;

[0012] S2. The meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast;

[0013] S3. Formulate flight plans under abnormal weather conditions based on the abnormal weather in the comprehensive forecast and obtain abnormal building construction images;

[0014] S4, using dark channel dehazing to improve the quality of abnormal construction images, and using SegNet deep network to extract key information of construction progress from enhanced abnormal construction images;

[0015] S5. Compare the key construction progress information with the actual progress information of the previous period, the planned progress information and the historical progress information of the same period, and comprehensively evaluate the progress deviation;

[0016] S6. Visually present the construction progress assessment results based on construction images through 3D modeling, and generate a progress assessment report containing images, data analysis and suggestions.

[0017] Optionally, the meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast, including the following steps:

[0018] S21, the meteorological measurement station collects meteorological parameter data, wherein the meteorological parameter data at least includes temperature data, humidity data, air pressure data, wind speed data and wind direction data;

[0019] S22, the meteorological measurement station collects meteorological images, wherein the meteorological images at least include cloud images, radar images and infrared images;

[0020] S23, extracting key features from meteorological parameter data and meteorological images;

[0021] S24, establish a time series model, input meteorological parameter data and key features of meteorological images, make a comprehensive forecast of future weather, and obtain the weather conditions for the next week;

[0022] S25. According to the predicted weather conditions, a threshold is set to classify normal weather and abnormal weather;

[0023] S26. If the weather condition is within the range of the set threshold, it is determined to be normal weather;

[0024] If it exceeds the set threshold, it is judged as abnormal weather.

[0025] Optionally, the meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast, and further includes the following steps:

[0026] If the comprehensive forecast result is normal weather, a flight plan under normal weather conditions will be formulated;

[0027] Acquire target building construction images and extract key information about the construction progress.

[0028] Optionally, the step of improving the quality of the abnormal building construction image by using dark channel dehazing, and extracting key information of the construction progress from the enhanced abnormal building construction image by using the SegNet deep network comprises the following steps:

[0029] S41, obtaining an original high-resolution abnormal building construction image from the abnormal building construction image;

[0030] S42, downsampling the original high-resolution abnormal building construction image to obtain a downsampled image with reduced resolution;

[0031] S43, calculating a dark channel on the downsampled image, establishing a minimum value table to accelerate the calculation, and estimating a preliminary transmittance of the downsampled image based on the dark channel result;

[0032] S44, performing morphological edge detection on the original high-resolution abnormal building construction image to obtain an edge binary map, and using the edge binary map to optimize the initial transmittance of the abnormal building construction image to obtain an optimized transmittance of the abnormal building construction image;

[0033] S45, measuring the ambient light intensity, and calculating a preliminary defogging image using a defogging model according to the ambient light, the optimized transmittance of the abnormal building construction image, and the original high-resolution abnormal building construction image;

[0034] S46, upsampling the preliminary defogging image to obtain a defogging image of the same size as the original high-resolution abnormal building construction image, and processing to obtain a final defogging image;

[0035] S47. Collect a dataset of abnormal building construction images at different construction stages and annotate key building element data;

[0036] S48. Use the labeled key building element data to train the SegNet network model;

[0037] S49. Use the trained SegNet network model to predict the final dehazed image and obtain the semantic segmentation results of the building elements;

[0038] S410: Evaluate current construction progress information based on the existence of different building elements in the segmentation result.

[0039] Optionally, the step of calculating a dark channel on the downsampled image, establishing a minimum value table to accelerate the calculation, and estimating a preliminary transmittance of the downsampled image according to the dark channel result comprises the following steps:

[0040] S431, normalizing the downsampled image and mapping it to a set range (0-1);

[0041] S432, sliding a window in the downsampled image, traversing all pixels in each window, and finding the minimum value of the red channel, the green channel, and the blue channel;

[0042] S433, taking the minimum value of the red channel, green channel, and blue channel of each window as the dark channel value of the window;

[0043] S434, merging the dark channel values ​​of the red channel, the green channel, and the blue channel to obtain a dark channel map of the downsampled image;

[0044] S435, selecting the pixel point with the lowest empirical value of brightness from the dark channel image (the empirical value is usually 0.1%);

[0045] S436, finding the brightness value corresponding to the empirical value pixel in the original downsampled image, taking the average value as the global atmospheric light value, and calculating the preliminary transmittance of the downsampled image;

[0046] The calculation formula of the preliminary transmittance of the downsampled image is:

[0047]

[0048] In the formula, ω represents the experience weight;

[0049] J represents the dark channel value;

[0050] A represents the global atmospheric value;

[0051] K represents the preliminary transmittance of the downsampled image.

[0052] Optionally, the step of performing morphological edge detection on the original high-resolution abnormal building construction image to obtain an edge binary map, and optimizing the initial transmittance of the abnormal building construction image using the edge binary map to obtain the optimized transmittance of the abnormal building construction image comprises the following steps:

[0053] S441, high-pass filtering is performed on the original high-resolution abnormal building construction image to filter out uniform areas and retain edges and details;

[0054] S442, performing Sobel edge detection on the filtered image to detect image edges;

[0055] S443, performing double thresholding processing on the Sobel edge detection result to obtain an edge binary image;

[0056] S444, performing expansion and erosion operations on the edge binary image to remove noise;

[0057] S445, using the preliminary transmittance of the down-sampled image as the initial transmittance of the abnormal building construction image;

[0058] S446, for each pixel point in the non-edge area, a weight is calculated according to the spatial distance between the pixel point and the neighboring pixel points in the preset neighborhood, and a weighted average of the transmittance in the neighborhood is calculated, and the transmittance of the pixel point is set to the calculated weighted average transmittance to adjust the transmittance of the non-edge area;

[0059] S447, repeat the step of S446 until the transmittance of all non-edge areas is smoothed to the average value in the neighborhood, and the optimized transmittance of the original image is obtained.

[0060] Optionally, for each pixel point in the non-edge area, a weight is calculated according to the spatial distance between the pixel point and the neighboring pixel points in a preset neighborhood, and a weighted average of the transmittance in the neighborhood is calculated, and the transmittance of the pixel point is set to the calculated weighted average transmittance to adjust the transmittance of the non-edge area, which includes the following steps:

[0061] S4461, define the neighborhood size, and for each pixel point in the non-edge area, traverse each pixel point in the defined neighborhood;

[0062] S4462, using Euclidean distance to calculate the spatial distance between the pixel point and each pixel point in the neighborhood;

[0063] S4463, setting the influence weight of each pixel, wherein the weight is inversely proportional to the distance;

[0064] S4464, for each neighborhood pixel, multiply the transmittance by the corresponding weight, then sum all the products to obtain the sum of the weighted transmittances, then calculate the sum of all weight values, divide the sum of the weighted transmittances by the sum of the weights, and calculate the weighted average transmittance of the pixels in the neighborhood;

[0065] S4465, setting the transmittance of the pixel point to the weighted average transmittance obtained in the previous step;

[0066] The formula for calculating the spatial distance between a pixel and each pixel in the neighborhood is as follows:

[0067]

[0068] Where (x1, y1) represents the coordinates of the current pixel;

[0069] (x2,y2) represents the coordinates of a pixel in the neighborhood.

[0070] Optionally, the measuring of the ambient light intensity and calculating the preliminary defogging image using a defogging model according to the ambient light, the optimized transmittance of the abnormal building construction image, and the original high-resolution abnormal building construction image comprises the following steps:

[0071] S451, in the original high-resolution abnormal building construction image, selecting the brightest empirical value pixel point of the image;

[0072] S452, calculating the average values ​​of the red channel, the green channel, and the blue channel of the brightest empirical value pixel as the global ambient light intensity;

[0073] S453, traversing all pixels according to the defogging formula, for each pixel, subtracting the ambient light intensity from the color value of the pixel point in the red channel, the green channel, and the blue channel to remove the influence of the ambient light on the image;

[0074] S454, adding the ambient light intensity back to the red channel, the green channel, and the blue channel to restore the color value of the pixel after defogging, and obtaining a preliminary defogging image.

[0075] Optionally, the step of using the trained SegNet network model to predict the final dehazed image to obtain a semantic segmentation result of the building elements comprises the following steps:

[0076] S491, normalizing the final defogging image after defogging;

[0077] S492, inputting the processed final dehazed image into the trained SegNet model for forward calculation;

[0078] S493, using the encoder of the SegNet model to extract features of the image, and the decoder upsamples the feature map to the original image size;

[0079] S494, while the decoder is upsampling, the feature map corresponding to the encoder is passed to the decoder through a skip connection, and then the output feature map is classified by softmax to obtain the probability of each pixel belonging to each category;

[0080] S495, classifying each pixel point according to the category with the highest probability, and generating a semantic segmentation result of the building element;

[0081] S496. Calculate the existence ratio and distribution information of various building elements according to the segmentation results;

[0082] S497. Visually compare the segmentation result with the original image and check the segmentation effect.

[0083] Optionally, the step of evaluating the current construction progress information according to the existence of different building elements in the segmentation result comprises the following steps:

[0084] S4101. Determine the correspondence between different building elements and building stages;

[0085] S4102. Counting the existence ratio and area proportion of different building elements in the segmentation results;

[0086] S4103, preliminarily determine the construction stage of the current building based on the statistical results;

[0087] S4104, comparing the determined current construction stage with the planned stage, if the current stage lags behind the planned stage, the progress is delayed;

[0088] If the current stage is ahead of the planned stage, there is an error in the progress and the quality is rechecked;

[0089] S4105. Determine the progress percentage of the current stage according to the completion status of different building elements, and give the evaluation result of the current construction progress by comprehensively considering the building elements, the proportion of corresponding engineering quantities, and the quality status;

[0090] S4106. Compare the evaluation results with historical data and planned progress to determine progress deviations.

[0091] Compared with the prior art, this application has the following beneficial effects:

[0092] 1. The present invention can collect meteorological parameter data and meteorological images in real time by setting up meteorological measurement stations and drones to monitor the meteorological conditions of construction sites, so that the latest weather information can be obtained in time to provide decision-making basis for construction site managers. By using meteorological parameter data and meteorological images, a time series model can be established and key features can be extracted to make a comprehensive forecast of the weather conditions in the coming week, so that weather planning can be done in advance and corresponding measures can be taken to deal with possible abnormal weather. By realizing real-time wireless communication between the meteorological measurement station and the drone control station, a drone flight plan under normal weather conditions can be formulated according to the comprehensive forecast results, thereby improving monitoring efficiency and reducing the waste of human resources.

[0093] 2. The present invention effectively improves the clarity and contrast of abnormal construction images through dark channel dehazing technology, making the details in the image more obvious, providing a better basis for subsequent analysis and processing. By using the deep learning network SegNet, key building elements can be accurately extracted from the dehazed image, providing accurate data support for construction progress assessment based on construction images.

[0094] 3. The present invention is not only applicable to conventional construction images, but can also process images under abnormal conditions, such as images under weather conditions such as haze, rain and fog. Through morphological edge detection and transmittance optimization, the image quality is further improved, making the architectural elements in the image more distinct. Not only is the image quality enhanced, but key construction progress information can also be extracted from the image, providing strong data support for the management and monitoring of construction.

[0095] 4. The present invention can accurately identify and analyze key information in construction images, provide more accurate construction progress assessment based on construction images, and through automated image processing and analysis processes, can obtain construction progress information in real time, and promptly discover progress deviations and problems. Compared with traditional manual assessments, the present invention is more objective and accurate, reducing the interference of human factors. The assessment method can be adjusted and optimized according to different construction projects and construction conditions to meet different needs. By comparing with historical data, it can predict future construction progress and possible problems, and provide forward-looking suggestions for construction management. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown:

[0097] Figure 1 The present invention is a flowchart of a construction progress assessment method based on construction images according to an embodiment of the present invention. DETAILED DESCRIPTION

[0098] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0099] According to an embodiment of the present invention, a construction progress assessment method based on construction images is provided.

[0100] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the construction image-based building progress assessment method according to an embodiment of the present invention, the construction image-based building progress assessment method comprises the following steps:

[0101] S1. Set up a meteorological measurement station at the construction site and equip the drone with a wireless communication module for the meteorological measurement station.

[0102] It should be explained that the site selection of meteorological measurement stations needs to consider the following factors: away from obstructions, with good ventilation conditions, and not affected by construction site operations. It is usually set up in an open high point. The meteorological element measuring instruments that need to be installed at the measurement station: thermometer, hygrometer, anemometer, precipitation meter, sunshine meter, etc. It is also equipped with data acquisition modules and transmission equipment. At the same time, it is necessary to establish a power supply system for the meteorological measurement station, which usually includes solar panels and storage batteries. The wireless communication module uses mobile communication modules such as GPRS, CDMA or wireless sensor networks of self-organizing networks to ensure real-time wireless communication with the drone control station. The meteorological station needs to have a certain computing and processing capability to store, process and transmit meteorological data. It is usually implemented by embedded computers or industrial computers. The meteorological station can also integrate video monitoring functions to obtain images through meteorological cameras to assist in judging weather conditions. It is necessary to formulate operation and maintenance procedures for meteorological stations, including instrument calibration, data maintenance, equipment repair and maintenance, etc.

[0103] S2. The meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast.

[0104] Preferably, the meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast, including the following steps:

[0105] S21, the meteorological measurement station collects meteorological parameter data, wherein the meteorological parameter data at least includes temperature data, humidity data, air pressure data, wind speed data and wind direction data;

[0106] S22, the meteorological measurement station collects meteorological images, wherein the meteorological images at least include cloud images, radar images and infrared images;

[0107] S23, extracting key features from meteorological parameter data and meteorological images;

[0108] S24, establish a time series model, input meteorological parameter data and key features of meteorological images, make a comprehensive forecast of future weather, and obtain the weather conditions for the next week;

[0109] S25. According to the predicted weather conditions, a threshold is set to classify normal weather and abnormal weather;

[0110] S26. If the weather condition is within the range of the set threshold, it is determined to be normal weather;

[0111] If it exceeds the set threshold, it is judged as abnormal weather.

[0112] Preferably, the meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast, and further includes the following steps:

[0113] If the comprehensive forecast result is normal weather, a flight plan under normal weather conditions will be formulated;

[0114] Acquire target building construction images and extract key information about the construction progress.

[0115] It needs to be explained that through time series analysis, meteorological parameter data can be used to establish a mathematical model to predict the changing trend within the next week; meteorological images can be used for image classification prediction through deep learning models to judge future weather conditions; multi-source data fusion of parameter data prediction and image prediction results can improve prediction accuracy; after predicting weather conditions, it is necessary to set reasonable thresholds to accurately identify abnormal weather that may affect flight; when formulating flight plans under normal weather, factors such as flight route, flight altitude, and camera parameters should be considered; when formulating flight plans under abnormal weather, factors such as lowering flight altitude and increasing flight stability should be considered; the flight control system can receive meteorological forecast results and automatically plan and generate the optimal flight plan under different weather conditions; real-time monitoring of meteorological changes during flight, update flight plans when necessary, and ensure flight safety; image processing is performed after the flight, and key construction progress information is extracted using technologies such as deep learning.

[0116] S3. Develop flight plans under abnormal weather conditions based on the abnormal weather in the comprehensive forecast and obtain abnormal building construction images.

[0117] It should be explained that abnormal weather mainly refers to meteorological conditions such as haze, strong winds, and sandstorms that seriously affect visibility and flight control. When formulating flight plans under abnormal weather, it is necessary to consider reducing the flight altitude, generally keeping it above the minimum safe altitude, reducing the flight speed, increasing flight stability, and preventing abnormal weather from affecting flight control. Shorten or adjust the flight route to avoid areas with strong winds. Install navigation aids such as barometers, radar altimeters, etc. to enhance flight safety. It is preferred to equip anti-fog mirrors, infrared cameras and other equipment to improve visibility. Increase the distance between drone waypoints to ensure a safe interval. Monitor the power, signal and other status of the drone, and return in time when the index approaches the limit value. Monitor meteorological changes in real time during flight and update the flight plan when necessary. Images acquired under abnormal weather need to be enhanced by defogging and denoising.

[0118] S4. Dark channel dehazing is used to improve the quality of abnormal construction images, and the SegNet deep network is used to extract key information of construction progress from the enhanced abnormal construction images.

[0119] Preferably, the method of improving the quality of abnormal building construction images by using dark channel dehazing and extracting key information of construction progress from the enhanced abnormal building construction images by using SegNet deep network comprises the following steps:

[0120] S41, obtaining an original high-resolution abnormal building construction image from the abnormal building construction image;

[0121] S42, downsampling the original high-resolution abnormal building construction image to obtain a downsampled image with reduced resolution;

[0122] S43, calculating a dark channel on the downsampled image, establishing a minimum value table to accelerate the calculation, and estimating a preliminary transmittance of the downsampled image based on the dark channel result;

[0123] S44, performing morphological edge detection on the original high-resolution abnormal building construction image to obtain an edge binary map, and using the edge binary map to optimize the initial transmittance of the abnormal building construction image to obtain an optimized transmittance of the abnormal building construction image;

[0124] S45, measuring the ambient light intensity, and calculating a preliminary defogging image using a defogging model according to the ambient light, the optimized transmittance of the abnormal building construction image, and the original high-resolution abnormal building construction image;

[0125] S46, upsampling the preliminary defogging image to obtain a defogging image of the same size as the original high-resolution abnormal building construction image, and processing to obtain a final defogging image;

[0126] S47. Collect a dataset of abnormal building construction images at different construction stages and annotate key building element data;

[0127] S48. Use the labeled key building element data to train the SegNet network model;

[0128] S49. Use the trained SegNet network model to predict the final dehazed image and obtain the semantic segmentation results of the building elements;

[0129] S410: Evaluate current construction progress information based on the existence of different building elements in the segmentation result.

[0130] In addition, examples are outlined below:

[0131] Suppose there is an abnormal building construction image, which is blurred due to factors such as fog and dust. In order to improve the image quality and extract key information about the construction progress, the following steps are taken:

[0132] Obtain the original high-resolution abnormal building construction image. Downsample the image to reduce the resolution to improve computational efficiency. Calculate the dark channel on the downsampled image, establish a minimum value table to accelerate the calculation, and estimate the preliminary transmittance based on the dark channel result. Perform morphological edge detection on the original high-resolution image to obtain an edge binary map, and use the edge binary map to optimize the initial transmittance. Measure the ambient light intensity, and use the defogging model to calculate the preliminary defogging image based on the ambient light, the optimized transmittance, and the original image. Upsample the preliminary defogging image to the original resolution and process it to obtain the final defogging image. Collect a dataset of abnormal building construction images at different construction stages and annotate key building element data. Use the annotated data to train the SegNet network model. Use the trained SegNet model to predict the final defogging image to obtain the semantic segmentation result of the building elements. Evaluate the current construction progress information based on the presence of different building elements in the segmentation result.

[0133] In the above example, key progress information is successfully extracted from the abnormal building construction image, ensuring accuracy in complex and challenging environments. Preferably, the dark channel is calculated on the downsampled image, a minimum value table is established to accelerate the calculation, and the preliminary transmittance of the downsampled image is estimated based on the dark channel result, including the following steps:

[0134] S431, normalizing the downsampled image and mapping it to a set range;

[0135] S432, sliding a window in the downsampled image, traversing all pixels in each window, and finding the minimum values ​​of the red channel, the green channel, and the blue channel (sliding window is a commonly used image processing technique, which is used to move and analyze the image with a certain step length. In this step, the minimum value of the three color channels (RGB) is found in each window);

[0136] S433, taking the minimum value of the red channel, green channel, and blue channel of each window as the dark channel value of the window (the concept of dark channel is based on the following observation: in most natural scenes, at least one color channel has a very small value, which is usually related to the shadow or dark part of the object surface. By taking the minimum value of the three channels in each window, the dark part information can be approximately found);

[0137] S434, merging the dark channel values ​​of the red channel, the green channel, and the blue channel to obtain a dark channel map of the downsampled image (combining the dark channel values ​​of each window to form a single dark channel map, which will be used for estimating the atmospheric light value and transmittance in subsequent steps);

[0138] S435, selecting the empirical value pixel with the lowest brightness from the dark channel image (it is generally believed that the darkest pixel in the dark channel image corresponds to the darkest part of the scene because the influence of atmospheric scattering or fog is minimal, and therefore it can be used as the empirical value pixel to estimate the atmospheric light value);

[0139] S436. Find the brightness values ​​corresponding to the empirical value pixels in the original downsampled image, take the average value as the global atmospheric light value, and calculate the preliminary transmittance of the downsampled image (by finding the pixel corresponding to the darkest pixel selected in the dark channel map in the original image, estimating the global atmospheric light value, which reflects the lighting conditions in the scene, and then using this atmospheric light value and the dark channel map, the preliminary transmittance can be calculated, which reflects the clarity or fog concentration of each pixel in the image);

[0140] The calculation formula of the preliminary transmittance of the downsampled image is:

[0141]

[0142] In the formula, ω represents the experience weight;

[0143] J represents the dark channel value;

[0144] A represents the global atmospheric value;

[0145] K represents the initial transmittance of the downsampled image. Preferably, the step of performing morphological edge detection on the original high-resolution abnormal building construction image to obtain an edge binary map, and optimizing the initial transmittance of the abnormal building construction image using the edge binary map to obtain the optimized transmittance of the abnormal building construction image comprises the following steps:

[0146] S441, high-pass filtering is performed on the original high-resolution abnormal construction image to filter out uniform areas and retain edges and details (the high-pass filter allows high-frequency signals (such as edges and details) to pass through, while suppressing low-frequency signals (such as uniform areas) and enhancing edge information in the image);

[0147] S442, performing Sobel edge detection on the filtered image to detect image edges (the Sobel operator is a common edge detection algorithm that detects edges by calculating the spatial gradient of image brightness. In the filtered image, edge information is more obvious, so that the Sobel operator can effectively detect edges);

[0148] S443, performing double thresholding processing on the Sobel edge detection result to obtain an edge binary image (double thresholding is an image segmentation technology that uses two thresholds to divide edge pixels into strong edges and weak edges);

[0149] S444, performing dilation and erosion operations on the edge binary image to remove noise (dilation and erosion are morphological operations used to smooth image edges, connect broken edge segments, and remove noise points);

[0150] S445, using the preliminary transmittance of the down-sampled image as the initial transmittance of the abnormal building construction image;

[0151] S446. For each pixel point in the non-edge area, a weight is calculated according to the spatial distance between the pixel point and the neighboring pixel points in the preset neighborhood, and a weighted average of the transmittance in the neighborhood is calculated, and the transmittance of the pixel point is set to the calculated weighted average transmittance to adjust the transmittance of the non-edge area (the transmittance is adjusted by taking a weighted average of the transmittance of each non-edge pixel point in its neighborhood to ensure that the transmittance is smoother in space and reduce the influence of local outliers);

[0152] S447, repeat the step of S446 until the transmittance of all non-edge areas is smoothed to the average value in the neighborhood, and the optimized transmittance of the original image is obtained.

[0153] Preferably, for each pixel point in the non-edge area, a weight is calculated according to the spatial distance between the pixel point and the neighboring pixel points in a preset neighborhood, and a weighted average of the transmittance in the neighborhood is calculated, and the transmittance of the pixel point is set to the calculated weighted average transmittance to adjust the transmittance of the non-edge area, which includes the following steps:

[0154] S4461, define the neighborhood size, and for each pixel point in the non-edge area, traverse each pixel point in the defined neighborhood;

[0155] S4462, using Euclidean distance to calculate the spatial distance between the pixel point and each pixel point in the neighborhood;

[0156] S4463, setting the influence weight of each pixel, wherein the weight is inversely proportional to the distance;

[0157] S4464, for each neighborhood pixel, multiply the transmittance by the corresponding weight, then sum all the products to obtain the sum of the weighted transmittances, then calculate the sum of all weight values, divide the sum of the weighted transmittances by the sum of the weights, and calculate the weighted average transmittance of the pixels in the neighborhood;

[0158] S4465, setting the transmittance of the pixel point to the weighted average transmittance obtained in the previous step;

[0159] The formula for calculating the spatial distance between a pixel and each pixel in the neighborhood is as follows:

[0160]

[0161] Where (x1, y1) represents the coordinates of the current pixel;

[0162] (x2, y2) represents the coordinates of a pixel point in the neighborhood. Preferably, the step of measuring the ambient light intensity and calculating the preliminary defogging image using a defogging model according to the ambient light, the optimized transmittance of the abnormal building construction image, and the original high-resolution abnormal building construction image comprises the following steps:

[0163] S451, in the original high-resolution abnormal building construction image, selecting the brightest empirical value pixel point of the image;

[0164] S452, calculating the average values ​​of the red channel, the green channel, and the blue channel of the brightest empirical value pixel as the global ambient light intensity;

[0165] S453, traversing all pixels according to the defogging formula, for each pixel, subtracting the ambient light intensity from the color value of the pixel point in the red channel, the green channel, and the blue channel to remove the influence of the ambient light on the image;

[0166] S454, adding the ambient light intensity back to the red channel, the green channel, and the blue channel to restore the color value of the pixel after defogging, and obtaining a preliminary defogging image.

[0167] Preferably, the method of using the trained SegNet network model to predict the final dehazed image to obtain the semantic segmentation result of the building elements comprises the following steps:

[0168] S491, normalizing the final defogging image after defogging (normalization is to eliminate the influence caused by scale differences between different images, and usually involves scaling the pixel values ​​of the image to a fixed range, such as 0 to 1);

[0169] S492, input the processed final dehazed image into the trained SegNet model for forward calculation (forward calculation refers to passing the input image through the network layer by layer until the last layer outputs the result);

[0170] S493, using the encoder of the SegNet model to extract features of the image, and the decoder upsamples the feature map to the original image size (the encoder gradually reduces the size of the image through convolutional layers and pooling layers while extracting image features, and the decoder gradually restores the image size through upsampling operations while combining feature maps of the corresponding levels in the encoder);

[0171] S494, while the decoder is upsampling, the feature map corresponding to the encoder is passed to the decoder through a skip connection, and then the output feature map is subjected to softmax classification to obtain the probability of each pixel belonging to each category (the skip connection allows the decoder to directly access the feature map of the early stage in the encoder, thereby retaining more detailed information, and the softmax classification converts the feature map into a probability distribution, and each pixel corresponds to a category probability);

[0172] S495, classifying each pixel point according to the category with the highest probability, and generating a semantic segmentation result of the building element;

[0173] S496. Calculate the existence ratio and distribution information of various building elements according to the segmentation results (by analyzing the segmentation results, the number and distribution of various building elements can be counted, and the information is used to evaluate the construction progress);

[0174] S497. Visually compare the segmentation result with the original image and check the segmentation effect.

[0175] Preferably, the step of evaluating the current construction progress information according to the existence of different building elements in the segmentation results comprises the following steps:

[0176] S4101. Determine the correspondence between different building elements and construction stages (associate building elements (such as foundation, wall, roof, etc.) with different stages of construction (such as piling, walling, capping, etc.));

[0177] S4102, calculating the existence ratio and area proportion of different building elements in the segmentation results (by analyzing the segmentation results, calculating the number of occurrences and the total area occupied by each building element in the image);

[0178] S4103. Preliminarily determine the construction stage of the current building based on the statistical results (by comparing the existence ratio and area proportion of different building elements and the corresponding relationship with the construction stage, the approximate construction stage of the building is inferred);

[0179] S4104, comparing the determined current construction stage with the planned stage, if the current stage lags behind the planned stage, the progress is delayed;

[0180] If the current stage is ahead of the planned stage, there is an error in the progress and the quality is rechecked;

[0181] S4105. Determine the progress percentage of the current stage according to the completion status of different building elements, and give the evaluation result of the current construction progress by comprehensively considering the building elements, the proportion of the corresponding engineering quantity, and the quality status (consider the complexity of the building elements, the size of the engineering quantity, and the quality standards, and comprehensively evaluate the current progress);

[0182] S4106. Compare the evaluation results with historical data and planned progress to determine progress deviations.

[0183] In addition, suppose there is a 3x3 neighborhood and a 5x5 image. If we want to traverse the neighborhood of the center pixel (coordinates (2, 2)), the neighborhood size is defined as 3x3, and all pixels in the 3x3 area around the center pixel will be considered.

[0184] The coordinates of the center pixel are (2, 2), and the corresponding value is 18;

[0185] Now we need to traverse the 3x3 neighborhood around the center pixel:

[0186] In this neighborhood, each pixel will be traversed and subsequent steps (such as calculating Euclidean distance, weight, etc.) will be performed.

[0187] For each neighborhood pixel, follow these steps:

[0188] Calculate the Euclidean distance between the current pixel and each pixel in the neighborhood. For example, for the center pixel (2, 2) and the pixel in the neighborhood (1, 1), the Euclidean distance is sqrt((2-1) 2 +(2-1) 2 )=sqrt(2), where sqrt means finding the square root of a number.

[0189] The influence weight of each pixel is calculated based on the Euclidean distance, and the weight is inversely proportional to the distance, that is, weight = 1 / distance.

[0190] Multiply the transmittance of each neighborhood pixel by its weight. For example, if a pixel in the neighborhood has a transmittance of 0.5 and a weight of 0.5, the product is 0.25.

[0191] All products are summed to get the sum of weighted transmittances. For example, if there are three neighborhood pixels whose transmittances multiplied by weights are 0.25, 0.5, and 0.25, respectively, then the sum is 1.

[0192] Calculate the sum of all weight values. For example, if the weights of three neighboring pixels are 0.5, 1, and 0.5 respectively, the sum is 2.

[0193] Divide the sum of the weighted transmittances by the sum of the weights to get the weighted average transmittance of the pixels in the neighborhood. For example, if the sum of the weighted transmittances is 1 and the sum of the weights is 2, the weighted average transmittance is 0.5.

[0194] The transmittance of the central pixel is set to the weighted average transmittance obtained in the above steps.

[0195] It should be explained that building images under abnormal weather conditions need to be preprocessed such as defogging to improve image quality before further analysis. Dark channel prior is an effective defogging method. Deep learning models such as SegNet are suitable for pixel-level semantic segmentation of building images and automatic identification of different building elements. A training set containing images of different building stages can be constructed to train the SegNet model to identify building elements. Engineering experience knowledge can correspond building elements to construction progress stages and evaluate the current stage. Information such as the proportion of building elements reflects the progress completion status, and the progress percentage can be determined in combination with engineering quantity calculation. Comparing the current progress with the planned progress and historical data can determine the progress deviation. The evaluation results are visualized and reports are generated, which are fed back to project managers to optimize planning and progress control. Compared with traditional methods, this technology realizes intelligent and automated building progress monitoring.

[0196] The SegNet network model is a semantic segmentation model based on a convolutional neural network (CNN) for pixel-level segmentation of images. It uses an encoder-decoder structure, where the encoder is used to extract image features and the decoder is used to map the features back to the original image size. The SegNet model performs well in many computer vision tasks, especially in image segmentation tasks.

[0197] Dark channel dehazing is a common image dehazing algorithm based on an observation: in natural images, any region of an opaque object contains at least one darker pixel, called a dark channel pixel. The dark channel dehazing algorithm takes advantage of this observation and removes haze from an image by estimating the dark channel pixels in the image as well as the global ambient light intensity.

[0198] S5. Compare the key construction progress information with the actual progress information of the previous period, the planned progress information and the historical progress information of the same period, and comprehensively evaluate the progress deviation.

[0199] It should be explained that by comparing the actual progress information, planned progress information and historical progress information of the previous period, the project's progress deviation can be comprehensively evaluated. The actual progress information is determined based on the actual amount of work completed and the time spent; the planned progress information is determined based on the schedule formulated in the project plan; the historical progress information is determined based on the actual progress of similar projects in the same period in the past. Comprehensive evaluation of progress deviation can be achieved by calculating deviation indicators. Common deviation indicators include progress deviation rate, progress deviation index and progress deviation degree. These indicators can help project managers understand the gap between the actual progress of the project and the planned progress, and provide a basis for formulating adjustment plans.

[0200] S6. Visually present the construction progress assessment results based on construction images through 3D modeling, and generate a progress assessment report containing images, data analysis and suggestions.

[0201] It should be explained that 3D modeling is a technology that uses computer software to create three-dimensional objects or scenes. It can generate realistic building models based on design drawings, data and parameters. In construction projects, professional modeling software can be used to create building models to reflect the actual situation of the project. Construction image-based construction progress assessment is the process of evaluating and analyzing the progress of the project during the construction project. By comparing the actual progress with the planned progress, the progress deviation of the project can be determined, and corresponding suggestions and measures can be put forward to ensure that the project is completed on time. By visually displaying the results of the construction progress assessment based on construction images, the actual situation of the project can be more intuitively understood. Through 3D modeling technology, the actual progress information and design drawings can be combined to generate a realistic building model. By visually presenting the model, the completion status of the building elements and the location of the progress deviation can be more clearly seen. The progress assessment report is a document generated based on the results of the construction progress assessment based on the construction image to provide the actual progress of the project, data analysis and suggestions. The report usually includes perspective screenshots of the building model, detailed data analysis of the progress deviation, and the completion status of key nodes. The report can help project managers better understand the progress of the project and make corresponding decisions and plans.

[0202] In summary, with the help of the above technical scheme of the present invention, the present invention can collect meteorological parameter data and meteorological images in real time by setting up meteorological measurement stations and drones to monitor the meteorological conditions of construction sites, so that the latest weather information can be obtained in time to provide decision-making basis for construction site managers. By using meteorological parameter data and meteorological images, by establishing a time series model and extracting key features, a comprehensive forecast of the weather conditions in the next week can be made, so that weather planning can be done in advance and corresponding measures can be taken to deal with possible abnormal weather. By realizing real-time wireless communication between the meteorological measurement station and the drone control station, a drone flight plan under normal weather conditions can be formulated according to the comprehensive forecast results, thereby improving monitoring efficiency and reducing waste of human resources. The present invention effectively improves the clarity and contrast of abnormal construction images through dark channel defogging technology, making the details in the image more obvious, and providing a better foundation for subsequent analysis and processing. By using the deep learning network SegNet, key building elements can be accurately extracted from the defogged image, providing a basis for construction based on construction images. The present invention provides accurate data support for the evaluation of construction progress. The present invention is not only applicable to conventional construction images, but also can process images under abnormal conditions, such as images under weather conditions such as haze, rain and fog. Through morphological edge detection and transmittance optimization, the image quality is further improved, making the architectural elements in the image more distinct. Not only is the image quality enhanced, but also key construction progress information can be extracted from the image, providing strong data support for the management and monitoring of construction. The present invention can accurately identify and analyze key information in construction images, provide more accurate construction progress evaluation based on construction images, and through automated image processing and analysis processes, can obtain construction progress information in real time, and timely discover progress deviations and problems. Compared with traditional manual evaluation, the present invention is more objective and accurate, reduces the interference of human factors, and can adjust and optimize the evaluation method according to different construction projects and construction conditions to meet different needs. By comparing with historical data, it can predict future construction progress and possible problems, and provide forward-looking suggestions for construction management.

[0203] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A construction progress assessment method based on construction images, characterized in that: The construction progress assessment method based on construction images comprises the following steps: S1. Set up a meteorological measurement station at the construction site and equip the drone with a wireless communication module for the meteorological measurement station; S2. The meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast; S3. Formulate flight plans under abnormal weather conditions based on the abnormal weather in the comprehensive forecast and obtain abnormal building construction images; S4, using dark channel dehazing to improve the quality of abnormal construction images, and using SegNet deep network to extract key information of construction progress from enhanced abnormal construction images; S5. Compare the key construction progress information with the actual progress information of the previous period, the planned progress information and the historical progress information of the same period, and comprehensively evaluate the progress deviation; S6. Visually present the construction progress assessment results based on the construction images through 3D modeling, and generate a progress assessment report containing images, data analysis and suggestions; The method of improving the quality of abnormal construction images by using dark channel dehazing and extracting key information of construction progress from the enhanced abnormal construction images by using SegNet deep network includes the following steps: S41, obtaining an original high-resolution abnormal building construction image from the abnormal building construction image; S42, downsampling the original high-resolution abnormal building construction image to obtain a downsampled image with reduced resolution; S43, calculating a dark channel on the downsampled image, establishing a minimum value table to accelerate the calculation, and estimating a preliminary transmittance of the downsampled image based on the dark channel result; S44, performing morphological edge detection on the original high-resolution abnormal building construction image to obtain an edge binary map, and using the edge binary map to optimize the initial transmittance of the abnormal building construction image to obtain an optimized transmittance of the abnormal building construction image; S45, measuring the ambient light intensity, and calculating a preliminary defogging image using a defogging model according to the ambient light, the optimized transmittance of the abnormal building construction image, and the original high-resolution abnormal building construction image; S46, upsampling the preliminary defogging image to obtain a defogging image of the same size as the original high-resolution abnormal building construction image, and processing to obtain a final defogging image; S47. Collect a dataset of abnormal building construction images at different construction stages and annotate key building element data; S48. Use the labeled key building element data to train the SegNet network model; S49. Use the trained SegNet network model to predict the final dehazed image and obtain the semantic segmentation results of the building elements; S410: Evaluate current construction progress information based on the existence of different building elements in the segmentation result.

2. A construction progress assessment method based on construction images according to claim 1, characterized in that: The meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast, including the following steps: S21, the meteorological measurement station collects meteorological parameter data, wherein the meteorological parameter data at least includes temperature data, humidity data, air pressure data, wind speed data and wind direction data; S22, the meteorological measurement station collects meteorological images, wherein the meteorological images at least include cloud images, radar images and infrared images; S23, extracting key features from meteorological parameter data and meteorological images; S24, establish a time series model, input meteorological parameter data and key features of meteorological images, make a comprehensive forecast of future weather, and obtain the weather conditions for the next week; S25. According to the predicted weather conditions, a threshold is set to classify normal weather and abnormal weather; S26. If the weather condition is within the range of the set threshold, it is determined to be normal weather; If it exceeds the set threshold, it is judged as abnormal weather.

3. A construction progress assessment method based on construction images according to claim 2, characterized in that: The meteorological measurement station predicts the weather conditions of the construction site in the next week, obtains meteorological parameter data and meteorological images, makes a comprehensive forecast of the future weather based on the meteorological parameter data and meteorological images, and divides the meteorological conditions into normal weather and abnormal weather based on the results of the comprehensive forecast, and further includes the following steps: If the comprehensive forecast result is normal weather, a flight plan under normal weather conditions will be formulated; Acquire target building construction images and extract key information about the construction progress.

4. The construction progress assessment method based on construction images according to claim 1, characterized in that: The step of calculating the dark channel on the downsampled image, establishing a minimum value table to accelerate the calculation, and estimating the preliminary transmittance of the downsampled image according to the dark channel result comprises the following steps: S431, normalizing the downsampled image and mapping it to a set range; S432, sliding a window in the downsampled image, traversing all pixels in each window, and finding the minimum value of the red channel, the green channel, and the blue channel; S433, taking the minimum value of the red channel, green channel, and blue channel of each window as the dark channel value of the window; S434, merging the dark channel values ​​of the red channel, the green channel, and the blue channel to obtain a dark channel map of the downsampled image; S435, selecting the pixel point with the lowest empirical value of brightness from the dark channel image; S436, finding the brightness value corresponding to the empirical value pixel in the original downsampled image, taking the average value as the global atmospheric light value, and calculating the preliminary transmittance of the downsampled image; The calculation formula of the preliminary transmittance of the downsampled image is: In the formula, ω represents the experience weight; J represents the dark channel value; A represents the global atmospheric value; K represents the preliminary transmittance of the downsampled image.

5. A construction progress assessment method based on construction images according to claim 4, characterized in that: The method of performing morphological edge detection on the original high-resolution abnormal building construction image to obtain an edge binary map, and optimizing the initial transmittance of the abnormal building construction image using the edge binary map to obtain the optimized transmittance of the abnormal building construction image comprises the following steps: S441, high-pass filtering is performed on the original high-resolution abnormal building construction image to filter out uniform areas and retain edges and details; S442, performing Sobel edge detection on the filtered image to detect image edges; S443, performing double thresholding processing on the Sobel edge detection result to obtain an edge binary image; S444, performing expansion and erosion operations on the edge binary image to remove noise; S445, using the preliminary transmittance of the down-sampled image as the initial transmittance of the abnormal building construction image; S446, for each pixel point in the non-edge area, a weight is calculated according to the spatial distance between the pixel point and the neighboring pixel points in the preset neighborhood, and a weighted average of the transmittance in the neighborhood is calculated, and the transmittance of the pixel point is set to the calculated weighted average transmittance to adjust the transmittance of the non-edge area; S447, repeat the step of S446 until the transmittance of all non-edge areas is smoothed to the average value in the neighborhood, and the optimized transmittance of the original image is obtained.

6. A construction progress assessment method based on construction images according to claim 5, characterized in that: The method of calculating the weight of each pixel point in the non-edge area according to the spatial distance between the pixel point and the neighboring pixel points in the preset neighborhood, calculating the weighted average transmittance in the neighborhood, and setting the transmittance of the pixel point to the calculated weighted average transmittance to adjust the transmittance of the non-edge area includes the following steps: S4461, define the neighborhood size, and for each pixel point in the non-edge area, traverse each pixel point in the defined neighborhood; S4462, using Euclidean distance to calculate the spatial distance between the pixel point and each pixel point in the neighborhood; S4463, setting the influence weight of each pixel, wherein the weight is inversely proportional to the distance; S4464, for each neighborhood pixel, multiply the transmittance by the corresponding weight, then sum all the products to obtain the sum of the weighted transmittances, then calculate the sum of all weight values, divide the sum of the weighted transmittances by the sum of the weights, and calculate the weighted average transmittance of the pixels in the neighborhood; S4465, setting the transmittance of the pixel point to the weighted average transmittance obtained in the previous step; The formula for calculating the spatial distance between a pixel and each pixel in the neighborhood is as follows: Where (x1, y1) represents the coordinates of the current pixel; (x2,y2) represents the coordinates of a pixel in the neighborhood.

7. A construction progress assessment method based on construction images according to claim 6, characterized in that: The measuring of the ambient light intensity and calculating the preliminary defogging image using the defogging model according to the ambient light, the optimized transmittance of the abnormal building construction image, and the original high-resolution abnormal building construction image include the following steps: S451, in the original high-resolution abnormal building construction image, selecting the brightest empirical value pixel point of the image; S452, calculating the average values ​​of the red channel, the green channel, and the blue channel of the brightest empirical value pixel as the global ambient light intensity; S453, traversing all pixels according to the defogging formula, for each pixel, subtracting the ambient light intensity from the color value of the pixel point in the red channel, the green channel, and the blue channel to remove the influence of the ambient light on the image; S454, adding the ambient light intensity back to the red channel, the green channel, and the blue channel to restore the color value of the pixel after defogging, and obtaining a preliminary defogging image.

8. The construction progress assessment method based on construction images according to claim 7 is characterized in that: The method of using the trained SegNet network model to predict the final dehazed image and obtain the semantic segmentation result of the building elements includes the following steps: S491, normalizing the final defogging image after defogging; S492, inputting the processed final dehazed image into the trained SegNet model for forward calculation; S493, using the encoder of the SegNet model to extract features of the image, and the decoder upsamples the feature map to the original image size; S494, while the decoder is upsampling, the feature map corresponding to the encoder is passed to the decoder through a skip connection, and then the output feature map is classified by softmax to obtain the probability of each pixel belonging to each category; S495, classifying each pixel point according to the category with the highest probability, and generating a semantic segmentation result of the building element; S496. Calculate the existence ratio and distribution information of various building elements according to the segmentation results; S497. Visually compare the segmentation result with the original image and check the segmentation effect.

9. A construction progress assessment method based on construction images according to claim 8, characterized in that: The step of evaluating the current construction progress information according to the existence of different building elements in the segmentation result includes the following steps: S4101. Determine the correspondence between different building elements and building stages; S4102. Counting the existence ratio and area proportion of different building elements in the segmentation results; S4103, preliminarily determine the construction stage of the current building based on the statistical results; S4104, comparing the determined current construction stage with the planned stage, if the current stage lags behind the planned stage, the progress is delayed; If the current stage is ahead of the planned stage, there is an error in the progress and the quality is rechecked; S4105. Determine the progress percentage of the current stage according to the completion status of different building elements, and give the evaluation result of the current construction progress by comprehensively considering the building elements, the proportion of corresponding engineering quantities, and the quality status; S4106. Compare the evaluation results with historical data and planned progress to determine progress deviations.

Images