Greenhouse safety risk early warning method, medium and equipment

Abstract

The invention relates to a greenhouse safety risk early warning method, medium and equipment, and the method comprises the steps: obtaining the current three-dimensional point cloud data and two-dimensional image of a greenhouse, comparing the three-dimensional point cloud data with priori initial data, obtaining the structure change data of the greenhouse, recognizing the two-dimensional image, obtaining the surface abnormal data, and carrying out the early warning of the safety risk of the greenhouse according to the structure change data and the surface abnormal data. And judging whether the greenhouse has a safety risk or not. The problems that in the prior art, a traditional steel structure detection method is low in efficiency, high in cost, difficult to discover hidden dangers, difficult to meet large-area high-precision detection requirements, insufficient in data processing capacity and the like are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of safety risk early warning, and in particular to a safety risk early warning method, medium and equipment for a greenhouse. Background Art

[0002] With the advancement of industrial modernization, steel structure greenhouses have been widely used due to their advantages such as efficient use of land and improved industrial production efficiency. However, steel structure greenhouses face a variety of safety risks during use, such as structural deformation, corrosion, connection failure, etc. These problems may lead to serious safety accidents and even endanger the safety of people's lives and property. Traditional steel structure inspection methods mainly rely on manual inspections, which are not only inefficient and costly, but also difficult to detect some hidden hidden dangers. In addition, as the first batch of steel structure greenhouses gradually approach their design life, it becomes particularly necessary to conduct regular safety assessments on them.

[0003] In the process of using the steel structure of the material yard greenhouse, the safety risk warning method faces many technical background problems. First, the traditional steel structure detection method mainly relies on manual regular inspections. This method is not only inefficient and costly, but also difficult to detect hidden dangers in hidden parts in time, resulting in a lag in the investigation of safety hazards. Secondly, the environment of the material yard greenhouse is complex, and the steel structure is exposed to harsh natural conditions for a long time. It is easily affected by factors such as corrosion, fatigue damage, and wind and snow loads, and the existing detection technology is difficult to monitor these dynamic changes in real time and comprehensively. In addition, with the expansion of the scale of the material yard and the increase in the complexity of the steel structure, the traditional monitoring method has been difficult to meet the needs of large-scale and high-precision detection. Finally, insufficient data processing and analysis capabilities are also a key issue. The existing monitoring system often lacks intelligent data analysis methods, making it difficult to quickly and accurately process the large amount of data collected, and thus unable to issue effective safety warnings in time. The existence of these problems has seriously restricted the development and application of safety risk warning technology for steel structures in material yard greenhouses.

[0004] Therefore, how to improve the traditional steel structure detection methods in the existing technology, such as low efficiency, high cost, difficulty in discovering hidden dangers, difficulty in meeting the needs of large-area high-precision detection, and insufficient data processing capabilities, is a technical problem that needs to be urgently solved in this field. Summary of the invention

[0005] Based on this, the purpose of this application is to provide a greenhouse safety risk warning method, medium and equipment to solve at least one technical problem mentioned in the above background technology.

[0006] In a first aspect, the present application provides a greenhouse safety risk early warning method, comprising:

[0007] Obtain the current 3D point cloud data and 2D images of the greenhouse;

[0008] Compare the three-dimensional point cloud data with the prior initial data to obtain the structural change data of the greenhouse.

[0009] Identify the two-dimensional image to obtain the surface anomaly data.

[0010] Judge whether there is a safety risk in the greenhouse according to the structural change data and the surface anomaly data.

[0011] Furthermore, the steps of judging whether there is a safety risk in the greenhouse according to the structural change data and the surface anomaly data include:

[0012] Construct and train a neural network model with the structural change data and the surface anomaly data as inputs and the safety risk coefficient as the output, which is the risk identification model; the structural change data includes any one or more of the point cloud difference data, deformation data, and displacement data.

[0013] Input the structural change data and the surface anomaly data into the risk identification model to obtain the safety risk coefficient.

[0014] Judge whether the safety risk coefficient is greater than the set coefficient threshold. If so, there is a safety risk in the greenhouse; if not, there is no safety risk in the greenhouse.

[0015] Furthermore, the steps of obtaining the point cloud difference data include:

[0016] Perform a pass-through filter on the three-dimensional point cloud data to segment and obtain several local point cloud data of the greenhouse; the local point cloud data includes any one or more of the column point cloud, arch point cloud, crossbeam point cloud, tie rod point cloud, and strut point cloud.

[0017] The prior initial data includes the initial point cloud data of each part of the greenhouse; compare each local point cloud data with the initial point cloud data to obtain the point cloud difference data of each part.

[0018] Furthermore, the steps of obtaining the deformation data and the displacement data include:

[0019] Perform line fitting on each local point cloud data to obtain the current line equation and obtain the current intersection points of each line equation.

[0020] The prior initial data also includes the prior line equation and the prior intersection points of the greenhouse.

[0021] Obtain the structural deformation data according to the current line equation and the prior line equation.

[0022] Obtain the structural displacement data according to the current intersection points and the prior intersection points.

[0023] Furthermore, the steps of obtaining the surface anomaly data include:

[0024] Input a two-dimensional image into a recognition model to obtain a number of surface abnormal regions;

[0025] Input each surface abnormal region into a semantic segmentation model to obtain the abnormal parameters of each surface abnormal region; the abnormal parameters include abnormal position, abnormal type, and abnormal size;

[0026] Statistically analyze the abnormal parameters of each surface abnormal region to obtain surface abnormal data.

[0027] Furthermore, the step of statistically analyzing the abnormal parameters of each surface abnormal region to obtain surface abnormal data includes:

[0028] Set a first safety factor for each surface abnormal region according to the abnormal position;

[0029] Set a second safety factor for each surface abnormal region according to the abnormal type;

[0030] Sum the abnormal sizes of each abnormal region according to the weights of the first safety factor and the second safety factor, and statistically obtain the total surface abnormal data of the greenhouse.

[0031] Furthermore, the structure of the risk recognition model includes:

[0032] An input layer for receiving point cloud difference data, deformation data, displacement data, and surface abnormal data;

[0033] A feature extraction layer connected to the input layer for performing dimensionality reduction operations and convolutional operations on the input data to obtain 4 feature maps;

[0034] A first fully connected layer connected to the pooling layer for converting all feature maps into vectors of several nodes, and then introducing non-linearity through the ReLU activation function;

[0035] A second fully connected layer: connected to the first fully connected layer for further compressing the vectors of several nodes output by the first fully connected layer, and through the fully connected relationship between neurons, further combining and transforming the features to obtain an initial safety risk coefficient;

[0036] An output layer connected to the fully connected layer for optimizing the initial safety risk coefficient using a set activation function to obtain a finally determined safety risk coefficient.

[0037] Furthermore, the structure of the feature extraction layer includes:

[0038] A first bottleneck layer for performing dimensionality reduction operations on the input data and retaining the data feature information to obtain initial dimensionality-reduced data;

[0039] The CBL convolutional layer, connected to the first bottleneck layer, is used to perform a convolutional operation on the initial dimension-reduced data to extract local features of the initial dimension-reduced data, obtaining 4 initial feature maps;

[0040] The second bottleneck layer, connected to the CBL convolutional layer, is used to perform a dimension reduction operation on each initial feature map and retain data feature information, obtaining 4 optimized feature maps;

[0041] The pooling layer, connected to the second bottleneck layer, is used to perform downsampling on each optimized feature map, reduce the spatial size of the optimized feature map, extract main features, so as to reduce the computational amount, obtaining 4 finally determined feature maps.

[0042] In a second aspect, the present application also provides a computer storage medium storing executable program code; the executable program code is used to execute the greenhouse safety risk warning method according to any one of the first aspect.

[0043] In a third aspect, the present application also provides a terminal device including a memory and a processor; the memory stores program code executable by the processor; the program code is used to execute the greenhouse safety risk warning method according to any one of the first aspect.

[0044] A greenhouse safety risk warning method, medium and device provided by the present invention, by acquiring the current 3D point cloud data and 2D image of the greenhouse, comparing the 3D point cloud data with prior initial data to obtain the structural change data of the greenhouse, so as to improve the detection efficiency and reduce the detection cost, then identifying the 2D image to obtain surface anomaly data, comprehensively detecting the hidden dangers on the surface of the greenhouse, realizing large-area high-precision detection, and finally judging whether there is a safety risk in the greenhouse according to the structural change data and surface anomaly data, improving the detection quota and data processing ability. It solves the problems in the prior art such as low efficiency, high cost, difficulty in discovering hidden dangers, difficulty in meeting the requirements of large-area high-precision detection, and insufficient data processing ability of traditional steel structure detection methods. Description of the Drawings

[0045] Figure 1 It is a flowchart of the greenhouse safety risk warning method according to an embodiment of the present invention;

[0046] Figure 2 It is a structural schematic diagram of the risk identification model according to an embodiment of the present invention;

[0047] Figure 3 It is a schematic flowchart of the greenhouse safety risk warning method according to an embodiment of the present invention. Detailed Embodiments

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] It should be noted that if there are directional indications in the embodiments of the present invention, such as up, down, left, right, front, back..., then the directional indications are only used to explain the relative position relationship and movement conditions between components in a specific posture. If the specific posture changes, the directional indications will also change accordingly. In addition, if there are descriptions such as "first, second", "S1, S2", "step one, step two" in the embodiments of the present invention, such descriptions are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features or indicating the execution order of the method, etc. Those skilled in the art can understand that all those that do not violate the invention points under the inventive concept of the invention should be included in the protection scope of the present invention.

[0050] Due to problems in the prior art such as low efficiency, high cost, difficulty in discovering hidden hazards, difficulty in meeting the requirements of large-area high-precision detection, and insufficient data processing capabilities in traditional steel structure detection methods, therefore, an optional safety risk warning process for a greenhouse is provided as Figures 1-3 shown. The present invention also provides a safety risk warning method for a greenhouse:

[0051] S1: Obtain the current three-dimensional point cloud data and two-dimensional image of the greenhouse;

[0052] Specifically, it is optional but not limited to using a collection device to surround the greenhouse to collect the three-dimensional point cloud data and two-dimensional image on the surface of the greenhouse; the collection device is optional but not limited to common scanning devices such as 3D line scan cameras, 3D structured light cameras, TOF cameras, laser scanners, etc. that can be used to obtain the three-dimensional point cloud data on the surface of an object and mobile phones, cameras, etc. that can be used to obtain the two-dimensional image on the surface of an object.

[0053] Preferably, since drones have the advantages of high flexibility, wide coverage, and low cost and can quickly complete large-area inspection tasks, therefore, it is optional to use a drone equipped with collection devices such as a high-definition camera, an infrared thermal imager, and a lidar to surround the greenhouse to achieve full-round monitoring of the greenhouse and obtain the current three-dimensional point cloud data and two-dimensional image of the greenhouse.

[0054] S2: Compare the three-dimensional point cloud data with the prior initial data to obtain the structural change data of the greenhouse;

[0055] Specifically, but not limited to, comparing the current 3D point cloud data of the greenhouse obtained in step S1 with the prior initial data to obtain the difference between the two, and then the structural change data of the greenhouse can be obtained; the structural change data includes any one or more of point cloud difference data, deformation data, and displacement data; the prior initial data includes the initial point cloud data of each part of the greenhouse; the initial point cloud data can be obtained by a person skilled in the art collecting point clouds after the installation of each part of the greenhouse structure, so as to obtain the initial point cloud data of the greenhouse.

[0056] Preferably, the step of comparing the 3D point cloud data with the prior initial data to obtain the structural change data of the greenhouse may include:

[0057] S21: Perform a pass-through filter on the 3D point cloud data to segment and obtain several local point cloud data of the greenhouse; the local point cloud data includes any one or more of column point clouds, arch frame point clouds, crossbeam point clouds, tie rod point clouds, and strut point clouds;

[0058] S22: The prior initial data includes the initial point cloud data of each part of the greenhouse; compare each local point cloud data with the initial point cloud data to obtain the point cloud difference data of each part

[0059] S23: Fit line segments to each local point cloud data to obtain the current line segment equation and obtain the current intersection points of each line segment equation;

[0060] S24: The prior initial data also includes the prior line segment equation and prior intersection points of the greenhouse;

[0061] Specifically, the prior initial data can be, under the normal structure of the greenhouse, taking a certain corner point of the greenhouse as the origin, determining the line segments where its columns, arch frames, crossbeams, tie rods, and struts are located as the prior line segment equations; taking the intersection points between each prior line segment as the prior intersection points under normal conditions.

[0062] S25: Obtain the structural deformation data according to the current line segment equation and the prior line segment equation;

[0063] Specifically, it can be determined, according to the current line segment equation and the prior line segment equation corresponding to the local point cloud data of each part of the greenhouse, the degree of deformation of each part of the greenhouse structure. For example, if the local point cloud corresponding to a certain position of the greenhouse only corresponds to one prior line segment, but in the current fitting situation, there are two current line segment equations, it indicates that there is a bend at this position of the greenhouse, and thus the structural deformation data can be obtained.

[0064] Exemplarily, taking the column point cloud as an example, each column point cloud corresponds to a prior straight-line equation. Optionally, the column point cloud can be segmented into several local point clouds, and the straight lines are respectively fitted, and then the vertical distances between the midpoints of each straight line are obtained. If the distance between the midpoints of adjacent straight lines is greater than the set distance threshold, then the adjacent two straight lines do not belong to the same straight-line equation, so as to obtain the number of straight-line equations corresponding to each column point cloud, and thus obtain the structural deformation data corresponding to the column point cloud.

[0065] S26: Obtain the structural displacement data according to the current intersection point and the prior intersection point.

[0066] Specifically, the intersection points include the intersection points inside the greenhouse and the corner points around. Optionally but not limited to, according to the current corner points around and the prior corner points around, obtain the overall displacement data of the greenhouse, including sinking data, collapse data, etc., and according to the current intersection points inside and the prior intersection points inside, obtain the surface depression data at various places on the surface of the greenhouse, etc. By integrating the overall displacement data and the surface depression data, the structural displacement data can be obtained.

[0067] S3: Identify the two-dimensional image to obtain the surface anomaly data;

[0068] Specifically, optionally but not limited to, construct and train an identification model and a semantic segmentation model, input the two-dimensional image into the identification model to obtain the surface anomaly regions contained in the image, and then input the surface anomaly regions into the semantic segmentation model in turn to obtain the anomaly parameters of each surface anomaly region.

[0069] Preferably, the step of identifying the two-dimensional image to obtain the surface anomaly data may optionally include:

[0070] S31: Input the two-dimensional image into the identification model to obtain several surface anomaly regions;

[0071] Specifically, the identification model can adopt any neural network model in the prior art; then, by batch collecting two-dimensional images of the same type of greenhouse and annotating the surface anomaly regions such as cracks and rust in the two-dimensional images, a training data set is constructed: then, according to the training data set, input the initial neural network model to train the parameters of the initial neural network model to obtain the trained neural network model as the identification model. It should be noted that this step is a preparatory work, which can be constructed and trained in advance, as long as it is constructed and trained well before using the neural network model for prediction.

[0072] Preferably, the identification model can adopt the yolo model.

[0073] S32: Input each surface anomaly region into the semantic segmentation model to obtain the anomaly parameters of each surface anomaly region; the anomaly parameters include the anomaly position, anomaly type, and anomaly size;

[0074] Specifically, the semantic segmentation model can adopt any neural network model in the prior art; obtain the training data set in step S31, annotate the abnormal parameters of each surface abnormal area, and construct a training data set: then, according to the training data set, input it into the initial neural network model to train the parameters of the initial neural network model, and obtain the trained neural network model as the semantic segmentation model. It should be noted that this step is a preparatory work, which can be pre-constructed and trained, as long as it is constructed and trained before the neural network model is needed for prediction. Preferably, the semantic segmentation model can adopt the SegFormer model.

[0075] Exemplarily, the two-dimensional image can be optionally divided into several local images, including the top surface area image, the upper side area image, the lower side area image, and the connection area image. When a surface abnormal area is recognized in each local image, its position is marked, and thus the abnormal position of each surface abnormal area can be obtained.

[0076] S33: Statistically calculate the abnormal parameters of each surface abnormal area to obtain surface abnormal data.

[0077] Specifically, optionally, according to step S31 and step S32, the abnormal parameters of each surface abnormal area in the two-dimensional image are obtained, and thus the abnormal parameters of each surface abnormal area are integrated to obtain surface abnormal data.

[0078] Preferably, the step of statistically calculating the abnormal parameters of each surface abnormal area to obtain surface abnormal data may optionally include:

[0079] S331: According to the abnormal position, set a first safety factor for each surface abnormal area;

[0080] S332: According to the abnormal type, set a second safety factor for each surface abnormal area;

[0081] S333: According to the first safety factor and the second safety factor, sum the abnormal sizes of each abnormal area according to the weight, and statistically obtain the total surface abnormal data of the greenhouse.

[0082] Specifically, optionally, according to the abnormal position and abnormal type of each surface abnormal area, the first safety factor and the second safety factor are set, and thus the abnormal sizes of each surface abnormal area after weighting are obtained according to the first safety factor and the second safety factor, and the total surface abnormal data of the greenhouse is statistically obtained; the abnormal position includes the top surface area, the upper side area, the lower side area, and the connection area; the abnormal type includes cracks, rust, etc.; the abnormal size includes the length of the abnormal area, the area of the abnormal area, etc.

[0083] Preferably, the first safety factor includes assigning a first weight coefficient to the surface anomaly area in the top surface area; assigning a second weight coefficient to the surface anomaly area in the upper side area; assigning a third weight coefficient to the surface anomaly area in the lower side area; assigning a fourth weight coefficient to the surface anomaly area in the connection area; the second safety factor includes assigning a fifth weight coefficient to the crack anomaly area; assigning a sixth weight coefficient to the rust anomaly area.

[0084] Exemplarily, taking the crack anomaly area in the top surface area as an example, if the abnormal size of this crack anomaly area is r, then the abnormal size after weighting for this crack anomaly area can be obtained according to Equation 3-1:

[0085] R = K1 * K5 * r 3-1

[0086] Wherein, R is the abnormal size after weighting for the crack anomaly area in the top surface area, K1 is the first weight coefficient, K5 is the fifth weight coefficient, and r is the abnormal size of the crack anomaly area in the top surface area.

[0087] Calculate the abnormal size after weighting for all abnormal areas in sequence and sum them up, then the total surface anomaly data of the greenhouse can be obtained.

[0088] S4: Judge whether there is a safety risk for the greenhouse according to the structural change data and the surface anomaly data.

[0089] Specifically, it is optional but not limited to constructing and training a neural network model with the structural change data and the surface anomaly data as inputs and the safety risk coefficient as the output, which is the risk identification model, and inputting the structural change data and the surface anomaly data into the risk identification model to obtain the greenhouse safety risk coefficient, so as to judge whether there is a safety risk for the greenhouse.

[0090] Preferably, the steps of judging whether there is a safety risk for the greenhouse according to the structural change data and the surface anomaly data include:

[0091] S41: Construct and train a neural network model with the structural change data and the surface anomaly data as inputs and the safety risk coefficient as the output, which is the risk identification model; the structural change data includes any one or more of point cloud difference data, deformation data, and displacement data;

[0092] Specifically, the neural network model can adopt any neural network model in the prior art; then, by batch collecting 3D point cloud data and 2D images of the same type of greenhouse, as well as corresponding structural change data and surface anomaly data, a training dataset is constructed: then, according to the training dataset, the initial neural network model is input to train various parameters of the initial neural network model to obtain the trained neural network model. It should be noted that this step is a preparatory work, which can be pre-constructed and trained, or can be carried out synchronously with step S1, as long as it is constructed and trained before the neural network model is needed for prediction.

[0093] Preferably, the structure of the risk identification model is as Figure 2 shown, including:

[0094] An input layer for receiving point cloud difference data, deformation data, displacement data, and surface anomaly data;

[0095] A feature extraction layer connected to the input layer for performing dimensionality reduction operations and convolutional operations on the input data to obtain 4 feature maps;

[0096] A first fully connected layer connected to the pooling layer for converting all feature maps into vectors of several nodes, and then introducing non-linearity through the ReLU activation function;

[0097] A second fully connected layer: connected to the first fully connected layer for further compressing the vectors of several nodes output by the first fully connected layer, and through the full connection relationship between neurons, further combining and transforming the features to obtain the initial safety risk coefficient;

[0098] An output layer connected to the fully connected layer for optimizing the initial safety risk coefficient using a set activation function to obtain the finally determined safety risk coefficient.

[0099] Further preferably, the structure of the feature extraction layer includes:

[0100] A first bottleneck layer for performing dimensionality reduction operations on the input data and retaining the data feature information to obtain the initial dimensionality reduction data;

[0101] A CBL convolutional layer connected to the first bottleneck layer for performing convolutional operations on the initial dimensionality reduction data to extract local features of the initial dimensionality reduction data to obtain 4 initial feature maps;

[0102] A second bottleneck layer connected to the CBL convolutional layer for performing dimensionality reduction operations on each initial feature map and retaining the data feature information to obtain 4 optimized feature maps;

[0103] The pooling layer, connected to the second bottleneck layer, is used to downsample each optimized feature map, reduce the spatial size of the optimized feature map, extract the main features, so as to reduce the computational amount and obtain 4 finally determined feature maps.

[0104] S42: Input the structure change data and surface anomaly data into the risk identification model to obtain the safety risk coefficient.

[0105] S43: Determine whether the safety risk coefficient is greater than the set coefficient threshold. If so, there is a safety risk in the greenhouse; if not, there is no safety risk in the greenhouse.

[0106] Specifically, it is optional but not limited to setting the coefficient threshold, and input the structure change data obtained in step S2 and the surface anomaly data obtained in step S3 into the risk identification model to obtain the safety risk coefficient of the current greenhouse, so as to determine whether the safety risk coefficient is greater than the set coefficient threshold. If so, there are risks such as collapse at the current position of the greenhouse, or the current structure of the greenhouse is severely deformed or corroded, etc., and it is not sufficient to support the greenhouse for a long time, so there is a safety risk; if not, there is no safety risk such as collapse or deformation in the greenhouse.

[0107] In this embodiment, a method for warning of safety risks of a greenhouse according to the present invention is given. By obtaining the current three-dimensional point cloud data and two-dimensional image of the greenhouse, the three-dimensional point cloud data is compared with the prior initial data to obtain the structure change data of the greenhouse, so as to improve the detection efficiency and reduce the detection cost. Then, the two-dimensional image is identified to obtain the surface anomaly data, comprehensively detect the hidden dangers on the surface of the greenhouse, and realize large-area high-precision detection. Finally, according to the structure change data and the surface anomaly data, it is judged whether there is a safety risk in the greenhouse, improving the detection accuracy and data processing ability. It solves the problems in the prior art such as low efficiency, high cost, difficulty in discovering hidden dangers, difficulty in meeting the requirements of large-area high-precision detection, and insufficient data processing ability in traditional steel structure detection methods.

[0108] On the other hand, the present invention also provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned methods for warning of safety risks of a greenhouse.

[0109] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores program code executable by the processor; the program code is used to execute any of the above-mentioned methods for warning of safety risks of a greenhouse.

[0110] Exemplarily, the program code can be segmented into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the program code in the terminal device.

[0111] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the terminal device may further include input / output devices, network access devices, a bus, etc.

[0112] The processor can be a Central Processing Unit (CPU), or can also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0113] The memory can be an internal storage unit of the terminal device, such as a hard disk or a memory. The memory can also be an external storage device of the terminal device, such as a plug-in hard disk equipped on the terminal device, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory can also include both the internal storage unit and the external storage device of the terminal device. The memory is used to store the program code and other programs and data required by the terminal device. The memory can also be used to temporarily store the data that has been output or will be output.

[0114] The above computer storage medium and terminal device are created based on the above-mentioned safety risk warning method for the greenhouse, and their technical functions and beneficial effects are not elaborated here. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0115] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A greenhouse safety risk early warning method, characterized in that: include: Obtain the current 3D point cloud data and 2D images of the greenhouse; Compare the 3D point cloud data with the prior initial data to obtain the structural change data of the greenhouse; Identify two-dimensional images and obtain surface abnormality data; Determine whether there are safety risks in the greenhouse based on the structural change data and surface abnormality data.

2. The method according to claim 1, characterized in that Based on the structural change data and surface abnormality data, the steps to determine whether the greenhouse has safety risks include: Construct and train a neural network model that takes structural change data and surface anomaly data as input and safety risk coefficient as output, which is a risk identification model; structural change data includes any one or more of point cloud difference data, deformation data, and displacement data; Input the structural change data and surface anomaly data into the risk identification model to obtain the safety risk coefficient; Determine whether the safety risk coefficient is greater than the set coefficient threshold. If so, there is a safety risk in the greenhouse. If not, there is no safety risk in the greenhouse.

3. The method according to claim 2, characterized in that The steps to obtain point cloud difference data include: Performing direct filtering on the three-dimensional point cloud data to segment and obtain a number of local point cloud data of the greenhouse; the local point cloud data includes any one or more of the column point cloud, the arch point cloud, the beam point cloud, the tie rod point cloud and the compression rod point cloud; The prior initial data includes the initial point cloud data of each part of the greenhouse; the point cloud data of each part is compared with the initial point cloud data to obtain the point cloud difference data of each part.

4. The method according to claim 3, characterized in that The steps of obtaining deformation data and displacement data include: Perform line segment fitting on each local point cloud data to obtain the current line segment equation and obtain the current intersection point of each line segment equation; A priori initial data also includes the a priori line segment equations and a priori intersection points of the greenhouse; According to the current line segment equation and the prior line segment equation, the structural deformation data is obtained; According to the current intersection point and the prior intersection point, the structural displacement data is obtained.

5. The method according to claim 2, characterized in that: The steps of obtaining surface anomaly data include: The two-dimensional image is input into the recognition model to obtain several surface abnormal areas; Each surface abnormal region is input into the semantic segmentation model to obtain abnormal parameters of each surface abnormal region; the abnormal parameters include abnormal position, abnormal type, and abnormal size; The abnormal parameters of each surface abnormal area are counted to obtain the surface abnormal data.

6. The method according to claim 5, characterized in that The steps of counting the abnormal parameters of each surface abnormal area to obtain surface abnormal data include: According to the abnormal position, a first safety factor is set for each surface abnormal area; According to the abnormality type, a second safety factor is set for each surface abnormality area; According to the first safety factor and the second safety factor, the abnormal size of each abnormal area is summed up according to the weight, and the total surface abnormal data of the greenhouse is obtained by statistics.

7. The method according to claim 2, characterized in that The structure of the risk identification model includes: The input layer is used to receive point cloud difference data, deformation data, displacement data and surface anomaly data; The feature extraction layer is connected to the input layer and is used to perform dimensionality reduction and convolution operations on the input data to obtain four feature maps; The first fully connected layer, connected to the pooling layer, is used to convert all feature maps into vectors of several nodes, and then introduce nonlinearity through the ReLU activation function; The second fully connected layer is connected to the first fully connected layer and is used to further compress the vectors of several nodes output by the first fully connected layer. Through the fully connected relationship between neurons, the features are further combined and transformed to obtain the initial safety risk coefficient. The output layer, connected to the fully connected layer, is used to optimize the initial safety risk factor using the set activation function to obtain the final safety risk factor.

8. The method according to claim 7, characterized in that The structure of the feature extraction layer includes: The first bottleneck layer is used to reduce the dimension of the input data and retain the data feature information to obtain the initial dimension-reduced data; The CBL convolution layer is connected to the first bottleneck layer and is used to perform a convolution operation on the initial dimension reduction data to extract local features of the initial dimension reduction data and obtain four initial feature maps; The second bottleneck layer is connected to the CBL convolution layer to reduce the dimension of each initial feature map and retain the data feature information to obtain 4 optimized feature maps; The pooling layer is connected to the second bottleneck layer and is used to downsample each optimized feature map, reduce the spatial size of the optimized feature map, extract the main features, and reduce the amount of calculation to obtain 4 finalized feature maps.

9. A computer storage medium, characterized in that An executable program code is stored; the executable program code is used to execute the greenhouse safety risk early warning method described in any one of claims 1-8.

10. A terminal device, characterized in that: It comprises a memory and a processor; the memory stores program codes executable by the processor; the program codes are used to execute the greenhouse safety risk warning method described in any one of claims 1-8.

Images