A landscaping engineering construction progress image detection method

Abstract

The application discloses a kind of garden greening engineering construction progress image detection methods, it is related to computer vision and image recognition technical field, the multiple-phase image of same garden scene is obtained by this method, after instance segmentation, image features are extracted and difference features are calculated, difference features are decomposed into topological change features representing construction structural changes and phenotype evolution features representing plant physiological gradual changes, and the two are independent in semantic space, the survival, survival and death state are determined by combining the phenotype features time sequence of individual plants, and the detection results are converted into the percentage of sub-project completion according to the unit conversion coefficient of the bill of quantities model, the application can decouple construction changes and plant growth changes at semantic level, improve the determination accuracy of key indicators such as survival rate, and output quantitative progress data that can be directly used for engineering settlement, suitable for the automatic image detection of garden greening engineering construction progress.

Description

Technical Field

[0001] This invention relates to the field of computer vision and image recognition technology, specifically a method for detecting construction progress images of landscaping projects. Background Technology

[0002] Currently, image-based automatic construction progress detection technology has been applied in the fields of architecture and civil engineering. Its basic principle is to collect images of the same construction site scene at different time points, and use image change detection, target detection, or comparison with building information models to identify newly added structural components, and then estimate the proportion of completed construction work. A similar existing technical solution is a construction progress detection method based on multi-temporal image comparison. This method periodically collects images of the construction site from fixed positions, registers the current temporal image with the reference temporal image, uses a convolutional neural network to extract features of the registered image and calculate feature differences, identifies the areas of change, and then associates and matches the changed areas with preset construction component categories to count the proportion of completed components to the total number of designed components, and outputs the construction progress percentage.

[0003] This existing technical solution is mainly designed for rigid building scenarios where the structure is relatively fixed and changes are primarily caused by human construction. Its change detection model implicitly assumes that areas outside of newly added construction components should maintain stable visual characteristics. When this solution is directly applied to landscaping projects, the non-rigid, naturally growing biological characteristics of plants in these scenarios easily lead to misinterpretations of natural growth phenomena such as leafing and color changes as scene changes, which are then filtered out as noise. This results in false alarms and missed alarms for construction progress milestones. Furthermore, this solution lacks a long-term physiological state tracking mechanism for individual objects, making it difficult to distinguish between the stress state of newly transplanted seedlings and their actual mortality state, leading to insufficient accuracy in calculating key indicators such as survival rate. In addition, this solution typically assumes a fixed image acquisition device location, making it difficult to adapt to different environments. In the context of garden construction sites with unpredictable camera positions and complex lighting and shadows, and the lack of a mechanism to map the recognition results to the bill of quantities model, the detection results are difficult to use directly for engineering measurement and settlement. Existing change detection technologies generally treat image difference features as an indivisible whole, and typically use threshold filtering or manual rule removal of dynamic noise regions to handle non-construction changes in the scene. There has long been a technical bias in this field that makes it difficult to effectively separate structural changes in construction and physiological changes in plants at the feature level. This invention, by constructing topological mapping branches and phenotypic mapping branches, and combining joint training of reconstruction loss and orthogonal constraint loss, achieves unsupervised decoupling of the two in the deep feature semantic space for the first time, breaking through the above-mentioned technical bias. Therefore, we propose a method for detecting the construction progress image of garden greening projects. Summary of the Invention

[0004] The purpose of this invention is to provide a method for detecting the construction progress image of a landscaping project.

[0005] The technical problem this invention aims to solve is: how to achieve automated and accurate image detection of the construction progress of landscaping projects without relying on 3D scanning equipment and using only ordinary RGB image sequences. Specifically, this includes: how to semantically decouple structural changes caused by human construction from gradual changes caused by natural plant growth in multi-temporal landscaping images, avoiding misjudging normal plant growth as construction anomalies and causing false alarms and missed alarms for construction progress nodes; how to accurately distinguish between the slow-release stress state and the actual death state of newly transplanted seedlings in images, avoiding misjudgments of key progress indicators such as survival rate due to visual feature confusion; and how to achieve adaptive alignment and anti-shadow fusion of cross-camera images under actual conditions such as non-fixed image acquisition equipment, inconsistent multi-camera perspectives, and complex and variable lighting and shadows, and accurately map the recognition results to the bill of quantities to achieve quantitative progress output that can be used for project settlement.

[0006] To address the aforementioned technical problems, this invention provides a method for detecting the construction progress image of a landscaping project, comprising: acquiring multi-temporal images of the same landscape scene taken at different time points, wherein the multi-temporal images include a temporal phase one image and a temporal phase two image; performing instance segmentation on the temporal phase one image and the temporal phase two image respectively to obtain masks for individual plants and hard structures in the landscape scene; inputting the temporal phase one image and the temporal phase two image respectively into a backbone network with shared weights to obtain a temporal phase one feature map and a temporal phase two feature map; calculating differential features based on the temporal phase one feature map and the temporal phase two feature map; inputting the differential features respectively into a topology mapping branch and a phenotypic mapping branch to obtain topology change features and phenotypic evolution features, wherein the topology change features are used to characterize... In a garden landscape, structural changes caused by construction activities are represented by phenotypic evolution features, which characterize the gradual changes caused by the physiological growth of plants. During network training, the topological change features and phenotypic evolution features are constrained to be independent of each other in the semantic space. Based on the topological change features, construction node categories are identified to obtain construction change detection results. Based on the phenotypic evolution features, combined with the phenotypic feature time sequence of each plant in multiple time phases, the physiological state of each plant is determined. The physiological state includes seedling establishment, survival, and death. According to the unit conversion coefficient in the preset bill of quantities model, the construction change detection results and the physiological state determination results of each plant are converted into standard engineering quantities to obtain the completion percentage of the sub-items of the garden greening project.

[0007] In one optional implementation, before performing instance segmentation on the Phase 1 image and the Phase 2 image respectively, the original multi-camera images are further subjected to color space conversion, adaptive gamma correction, feature point matching, homography matrix estimation, and brightness weighted fusion processing of overlapping areas to eliminate the influence of multi-camera perspective differences and shadow interference on the recognition results. In another optional implementation, the process of determining the physiological state of each individual plant uses a long short-term memory network to encode the temporal phenotypic features, and combines them with the standard seedling recovery time-series curve template in the preset seedling recovery phenotypic knowledge base. The dynamic time warping algorithm is used to calculate the matching distance between the phenotypic evolution coding sequence and the standard template, thereby determining whether the plant is in the seedling recovery, survival, or death state.

[0008] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0009] 1. This invention decomposes the differential features into two orthogonal subspaces: topological change features and phenotypic evolution features, and applies orthogonal constraints to them. This achieves effective decoupling of structural changes caused by human construction and gradual changes caused by natural plant growth at the feature semantic level, thereby reducing the false alarm and false alarm rates of construction progress nodes.

[0010] 2. This invention uses a long short-term memory network combined with a dynamic time warping algorithm to match the standard curve template in the seedling phenotypic knowledge base, which can distinguish between the seedling stress state and the actual death state in the time dimension, thus improving the calculation accuracy of key progress indicators such as survival rate and greening rate.

[0011] 3. This invention improves the adaptability of image recognition in garden site environments with variable camera positions and complex lighting and shadows by using anti-shadow processing based on adaptive calculation of gamma correction coefficients based on the mean brightness and multi-camera image alignment and fusion based on the homography matrix. Furthermore, by converting and mapping the detection results according to the conversion coefficients of the bill of quantities model, it achieves automatic correspondence between the recognition results and the completion percentage of sub-projects, enabling the detection results to be directly used for progress control and settlement of garden greening projects. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall process in an embodiment of the present invention;

[0013] Figure 2 This is a schematic diagram of the image preprocessing subdivision process in an embodiment of the present invention;

[0014] Figure 3 This is a schematic diagram of the detailed process for determining the physiological state of plants in an embodiment of the present invention. Detailed Implementation

[0015] The following is in conjunction with the appendix Figure 1-3The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0016] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0017] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. The method of the present invention can be executed by a processor deployed on a cloud processing platform or a local server. Its input is a sequence of RGB images of the garden site acquired by multi-source acquisition terminals such as smartphones, drones and fixed or mobile monitoring cameras, and its output is a report on the percentage completion of sub-projects that can be used for project settlement.

[0018] In the image acquisition and preprocessing stage, the processor acquires a set of original images of the garden site captured by multiple non-fixed-position acquisition terminals, denoted as... ,in Indicates the first The serial number of each acquisition station is determined by the complex lighting conditions and variable shadows at the garden construction site, and the fact that the acquisition terminal positions are not fixed. Therefore, the processor first performs anti-shadow processing on the original image: [The original image is then processed...] Convert the image from RGB color space to HSV color space, extract its luminance channel image, and calculate the normalized pixel mean of the luminance channel, denoted as . The gamma correction coefficient is adaptively calculated using the following formula:

[0019] ;

[0020] In the formula, Indicates the adaptive gamma correction coefficient. This represents the normalized pixel mean value of the luminance channel, and its value ranges from zero to one. This represents the natural logarithm operation. Using this method, when the overall scene is dark, the normalized mean is smaller, and the calculated gamma coefficient deviates more from one, thus achieving stronger non-linear brightening of dark areas and suppressing the interference of shadows on subsequent recognition. Then, non-linear correction is performed on the brightness channel using the following formula:

[0021] ;

[0022] In the formula, This represents the normalized brightness pixel value before correction. This represents the corrected, normalized brightness pixel value. For the gamma coefficients obtained above, after correction, the processed luminance channel is merged with the original chroma and saturation channels and converted back to RGB space to obtain an anti-shadow enhanced image;

[0023] After completing the anti-shadow processing, the processor extracts local feature points and their descriptors from anti-shadow enhanced images of adjacent camera positions or adjacent time phases, and performs feature point matching between image pairs to obtain a set of matched point pairs. Based on this set of matched point pairs, the homography matrix between images is estimated using the random sample consensus algorithm. The source image pixel coordinates are mapped to a unified reference coordinate system using the following formula:

[0024] ;

[0025] In the formula, Represents the homogeneous coordinates of pixels in the source image. This represents the homogeneous pixel coordinates after mapping to the reference coordinate system. The homography transformation matrix is ​​a 3x3 matrix. The processor maps the images of each camera position to the same reference coordinate system and fuses them by weighting the pixel brightness of the overlapping area. The output is a standardized aligned image of the unified scene, which is used as the temporal phase one image and temporal phase two image by the subsequent instance segmentation and feature decoupling module.

[0026] In the instance segmentation stage, the processor uses a pre-trained improved MaskR-CNN network to perform individual-level instance segmentation on the aligned Phase 1 and Phase 2 images, respectively, to obtain the mask and position index of each independent individual plant and hard structure. Based on the position index and appearance features, the masks of the same individual plant in different phase images are associated and matched to obtain the multi-phase mask sequence of the individual plant, which is used by the subsequent time-series physiological state determination module.

[0027] In the spatiotemporal bi-branch orthogonal eigenvalue decomposition stage, the processor aligns adjacent temporal images of the same scene. and Inputting the shared weights into the backbone network yields the corresponding feature maps. and The backbone network can be a visual Transformer network or a convolutional neural network with multi-scale feature extraction capabilities. The processor calculates the difference features of the two temporal feature maps using the following formula:

[0028] ;

[0029] In the formula, This represents the difference characteristic between phase one and phase two. , These represent the feature maps extracted from the backbone network of the phase 1 and phase 2 images, respectively. The processor will then perform differential feature extraction. By inputting the topology mapping branch and the phenotypic mapping branch respectively, the topology change features are obtained. Phenotypic evolutionary characteristics The two satisfy the following reconstruction relationship:

[0030] ;

[0031] In the formula, This represents the characteristic components in the topological change space, corresponding to construction-related changes such as hard paving and new planting of seedlings. The decomposition represents the feature components in the phenotypic evolution space, corresponding to physiological changes in plants such as crown growth and color gradation. To make the above decomposition semantically interpretable, a reconstruction loss is constructed during the network training phase to constrain the sum of the two branch outputs to approximate the original difference features.

[0032] ;

[0033] In the formula, Represents the reconstruction loss, symbol The L2 norm of a vector is used to construct an orthogonal constraint loss, which forces topological change features and phenotypic evolution features to be independent in the semantic space.

[0034] ;

[0035] In the formula, This represents the orthogonal constraint loss. This represents the transpose of the topological change eigenvector. To prevent the introduction of extremely small positive numbers due to a denominator of zero, the processor combines downstream task tags, including construction category tags and growth indicator tags, to construct a task monitoring loss. The total loss function is obtained by weighted summing of the three losses as follows, and used for end-to-end network training:

[0036] ;

[0037] In the formula, Indicates the total loss. , represents the weighting coefficients of orthogonal constraint loss and task supervision loss, respectively, which are preset hyperparameters. After the network training converges, during the inference stage, only the adjacent temporal images to be detected need to be input into the network to obtain the construction node detection results corresponding to the topological change features and the plant growth state regression results corresponding to the phenotypic evolution features. The two are output to the subsequent construction progress statistics module and the seedling survival determination module, respectively.

[0038] During the seedling establishment and survival time determination stage, the processor extracts leaf color histogram features, texture gradient features, and canopy porosity features from the multi-temporal mask sequence for each individual plant, arranging them in chronological order of shooting time to form the temporal feature sequence of that plant, denoted as... ,in Indicates the first The processor will process the phenotypic feature vectors of each phase into a time-series feature sequence. The input is fed into a Long Short-Term Memory (LSTM) network to obtain phenotypic evolution coding sequences with temporal context information, which are used to characterize the decay and recovery trends of the plant's phenotypic features over time. The processor pre-constructs a seedling recovery phenotypic knowledge base, which contains multiple standard seedling recovery time-series curve templates. Each template reflects the pattern of phenotypic characteristics of healthy seedlings first decreasing and then recovering during the normal seedling establishment period. The processor calculates the phenotypic evolution coding sequence. With each standard template in the knowledge base The cumulative cost matrix of the dynamic time-warped distance between them is recursively calculated using the following formula:

[0039] ;

[0040] In the formula, Represents a sequence forward Elements and sequences forward The cumulative matching cost between elements Representing the eigenvector and The Euclidean distance between them This indicates that the lowest value among the three is used to achieve non-linear scaling alignment of the two sequences on the time axis, and the lower right element of the cumulative cost matrix is ​​taken as the dynamic time warped distance between the two sequences:

[0041] ;

[0042] In the formula, Represents a sequence and The dynamic time-warped distance between them , Representing sequences respectively and The processor iterates through all standard templates in the knowledge base to obtain multiple dynamic time-warped distances. It selects the template corresponding to the shortest value as the target template and uses the shortest distance as the target distance. If the target distance is less than a preset threshold, and the phenotypic characteristics of the plant in its latest phase have recovered to the preset range of healthy characteristics, the plant is considered alive. If the target distance is less than the preset threshold but the latest phase characteristics are still in the decay range, the plant is considered slow to recover. If the plant's temporal characteristic sequence continues to decay and shows no recovery trend within the preset observation period, the plant is considered dead. If the dynamic time-warped distance between the plant's phenotypic evolution coding sequence and all standard templates is not less than the preset threshold, the plant is also considered dead. The processor counts the number of plants judged to be alive and the total number of designed plants, and calculates the survival rate using the following formula:

[0043] ;

[0044] In the formula, This indicates the survival rate of the greening. This indicates the number of plants that have been determined to be alive. Indicates the total number of plants in the design;

[0045] The preset threshold is determined based on the average intraclass dynamic time regularization distance between standard templates in the seedling recovery phenotypic knowledge base, and is generally 1.2 to 1.5 times the average distance. The range of health feature values ​​is determined based on the statistical mean ± 2 standard deviation of the leaf color histogram, texture gradient and canopy porosity features of healthy adult plants in the same batch. The preset observation period is set according to the physiological cycle of seedling recovery of plant species, and is generally 15 to 45 days after transplanting.

[0046] During the progress quantification mapping stage, the processor identifies the construction-related change areas from the topology change branch, such as newly added hard paving areas and new planting points. It then performs area and quantity statistics, and combines this with the aforementioned homography matrix to convert the pixel area into the actual ground area, obtaining the actual area and quantity of various identified objects. The processor reads the preset conversion coefficients for each sub-item of the bill of quantities model and converts the area or quantity of the identified change areas into standard quantities using the following formula:

[0047] ;

[0048] In the formula, This represents the converted standard quantity of work. Indicates the first The actual area or number of objects in the class identification. This indicates the preset unit conversion factor for this type of object in the bill of quantities model. The processor divides the calculated standard engineering quantity by the total design engineering quantity of the corresponding sub-item in the bill of quantities to obtain the completion percentage of the sub-item. It then combines the survival rate of the greening to correct the sub-items related to the determination of plant survival. Finally, it summarizes the completion percentage of each sub-item, generates a visual construction progress report, and outputs it to the construction supervision management platform.

[0049] To further illustrate the implementation of this invention, a landscaping project is used as an example: The project is designed with a total of 1,200 plants, including trees, shrubs, and ground cover, and includes hard paving areas and newly planted flower beds. Starting from the day the seedlings are transplanted, the construction team uses handheld terminals and fixed monitoring cameras to capture images of the site every few days, forming a multi-temporal image sequence. After processing by the method described in this invention, the system performs multi-camera alignment and anti-shadow fusion on each temporal image, eliminating interference from different shooting angles and tree canopy shadows. After segmentation using an improved MaskR-CNN network, independent masks are obtained for each plant and hard structure. After processing by a spatiotemporal bi-branch orthogonal feature decomposition network, the newly paved hard paving areas and newly added planting points are accurately identified as construction nodes corresponding to topological change features, while some trees... The gradual color change of leaves due to seasonal changes was classified as a phenotypic evolution feature and was not misjudged as a construction anomaly. For several trees in the early stage of transplanting with temporarily yellowing leaves, after long short-term memory network encoding and dynamic time warping matching, their phenotypic evolution curves matched the standard seedling recovery template in the seedling recovery phenotypic knowledge base with a high degree of matching and the distance was less than the preset threshold. Therefore, they were judged as seedlings in the recovery stage rather than dead, avoiding misjudgment of survival rate. As time went by, the phenotypic characteristics of this batch of trees gradually recovered to the healthy range, and the system updated their status to survival accordingly. Finally, based on the unit conversion coefficient of each item in the bill of quantities model, the system converted the identified hard paving area, number of new plants and survival rate into the corresponding percentage of completion of the sub-item project and generated a visual progress report for use by project supervisors and settlement personnel.

[0050] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for detecting the construction progress image of a landscaping project, characterized in that, include: Multi-temporal images of the same garden scene taken at different time points are acquired. These multi-temporal images include a first-phase image and a second-phase image. Instance segmentation is performed on both the first-phase and second-phase images to obtain masks for individual plants and hard structures within the garden scene. The first-phase and second-phase images are then input into a backbone network with shared weights to obtain first-phase feature maps and second-phase feature maps. Differential features are calculated based on these two feature maps and input into a topology mapping branch and a phenotypic mapping branch, respectively, to obtain topological change features and phenotypic evolution features. The topological change features are used to characterize structural changes in the garden scene caused by construction activities. The phenotypic evolution features are used to characterize the gradual changes caused by the plant's own physiological growth. During the network training phase, the topological change features and the phenotypic evolution features are constrained to be independent of each other in the semantic space. Based on the topological change features, the construction node categories are identified, and the construction change detection results are obtained. Based on the phenotypic evolution features, combined with the phenotypic feature time sequence of each plant in multiple time phases, the physiological state of each plant is determined. The physiological state includes seedling establishment, survival, and death. According to the unit conversion coefficient in the preset bill of quantities model, the construction change detection results and the physiological state determination results of each plant are converted into standard engineering quantities to obtain the completion percentage of the sub-items of the landscaping project.

2. The method for detecting the construction progress image of a landscaping project according to claim 1, characterized in that: The step of constraining the topology change features and the phenotypic evolution features to be independent of each other in the semantic space during the network training phase includes: constructing a reconstruction loss, which is used to constrain the sum of the topology change features and the phenotypic evolution features to approximate the difference features; constructing an orthogonal constraint loss, which is determined based on the ratio between the absolute value of the inner product of the topology change features and the phenotypic evolution features and the product of the L2 norm of the topology change features and the L2 norm of the phenotypic evolution features; constructing a task supervision loss by combining the labels corresponding to the construction node categories and the labels corresponding to the physiological states of each individual plant; and weighting and summing the reconstruction loss, the orthogonal constraint loss, and the task supervision loss to obtain the total loss used for network training.

3. The method for detecting construction progress images of landscaping projects according to claim 1, characterized in that: Based on the phenotypic evolution characteristics, and combined with the phenotypic feature time sequence of each individual plant at multiple time phases, the physiological state of each individual plant is determined. This includes: for each individual plant, extracting leaf color histogram features, texture gradient features, and canopy porosity features at each time phase, arranging them in chronological order of shooting time to obtain the temporal feature sequence of that individual plant; inputting the temporal feature sequence into a long short-term memory network to obtain the phenotypic evolution coding sequence of that individual plant; obtaining a pre-constructed seedling establishment phenotypic knowledge base, which stores multiple standard seedling establishment time-series curve templates; calculating the dynamic time warping distance between the phenotypic evolution coding sequence and each of the standard seedling establishment time-series curve templates; and determining the shortest dynamic time warping distance as the target distance. The target template is determined by the standard seedling recovery timeline template corresponding to the distance. If the target distance is less than a preset threshold and the phenotypic characteristics of the latest phase of the individual plant have recovered to the preset health characteristic value range, the physiological state of the individual plant is determined to be survival. If the target distance is less than the preset threshold and the phenotypic characteristics of the latest phase of the individual plant are in the decay range, the physiological state of the individual plant is determined to be seedling recovery. If the timeline characteristic sequence of the individual plant continues to decay and no recovery trend is observed within the preset observation period, the physiological state of the individual plant is determined to be death. If the dynamic time warping distance between the phenotypic evolution coding sequence of the individual plant and all the standard seedling recovery timeline templates is not less than the preset threshold, the physiological state of the individual plant is determined to be death.

4. The method for detecting construction progress images of landscaping projects according to claim 1, characterized in that: Before performing instance segmentation on the Phase 1 image and the Phase 2 image respectively, the method further includes: acquiring original images of the garden scene captured by multiple non-fixed camera acquisition terminals; converting each original image from RGB color space to HSV color space to obtain the luminance channel image of each original image; adaptively calculating gamma correction coefficients based on the normalized mean of the luminance channel image; performing nonlinear correction on the luminance channel image based on the gamma correction coefficients to obtain an anti-shadow enhancement image; extracting feature points and their descriptors from the anti-shadow enhancement images of adjacent camera positions respectively; performing feature point matching between image pairs to obtain matching point pairs; estimating the homography matrix between images based on the matching point pairs using a random sampling consensus algorithm; mapping each anti-shadow enhancement image to a unified reference coordinate system based on the homography matrix to obtain an aligned image; fusing each aligned image mapped to the reference coordinate system according to the pixel brightness weighting of the overlapping area to obtain a standardized scene image after alignment and fusion; and using the standardized scene image as the Phase 1 image and the Phase 2 image.

5. The method for detecting construction progress images of landscaping projects according to claim 4, characterized in that: The step of adaptively calculating the gamma correction coefficient based on the normalized mean of the luminance channel image includes: obtaining the gamma correction coefficient by the ratio of the natural logarithm of the normalized mean of the luminance channel image to the natural logarithm of 0.

5.

6. The method for detecting the construction progress image of a landscaping project according to claim 1, characterized in that: Instance segmentation is performed on the Phase 1 image and the Phase 2 image respectively, including: using a pre-trained MaskR-CNN network to perform individual-level instance segmentation on the Phase 1 image and the Phase 2 image respectively, to obtain the masks and position indices of each individual plant and hard structure; based on the position index and appearance features, the masks of the same individual plant in the Phase 1 image and the Phase 2 image are associated and matched to obtain the multi-temporal mask sequence of the individual plant.

7. The method for detecting construction progress images of landscaping projects according to claim 1, characterized in that: Based on the unit conversion coefficient in the preset bill of quantities model, the results of the construction change detection and the physiological status judgment results of each individual plant are converted into standard engineering quantities to obtain the completion percentage of the sub-items of the landscaping project. This includes: statistically analyzing the changed areas in the construction change detection results to obtain the actual area and actual quantity of each type of identified object; converting the actual area and actual quantity of each type of identified object into standard engineering quantities based on the unit conversion coefficient preset in the bill of quantities model; dividing the standard engineering quantities by the total design engineering quantity of the corresponding sub-item in the bill of quantities model to obtain the completion percentage of the sub-item; calculating the ratio of the number of individual plants judged to be in a surviving state to the total number of plants in the design to obtain the survival rate; and correcting the completion percentage of the sub-items involving plant survival judgment based on the survival rate.

8. The method for detecting the construction progress image of a landscaping project according to claim 1, characterized in that: The backbone network is selected from either the visual Transformer network or the convolutional neural network.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for detecting the construction progress image of landscaping projects as described in any one of claims 1 to 8.

Images