Construction site obstacle determination method and system based on image processing

Abstract

The invention discloses a construction site obstacle determination method and system based on image processing, and relates to the technical field of image feature analysis, and the method comprises the steps: obtaining a construction site obstacle image set; the method comprises the following steps: preprocessing a construction site obstacle image set to remove noise in the image, performing feature enhancement on the image to improve the identification degree of image features, performing rotation and color conversion on the first obstacle image set to enlarge the first obstacle image set, and performing color conversion on the second obstacle image set to improve the identification degree of the image features. And performing feature recognition on the third obstacle image set through the convolutional neural network model to determine the type of the to-be-verified obstacle, and finally verifying the type of the to-be-verified obstacle through the third obstacle image set to determine the precision of the convolutional neural network model, and if the precision of the convolutional neural network model meets the standard, putting the convolutional neural network model into use. The image is identified through the convolutional neural network model, so that the working intensity of identification personnel can be reduced, and meanwhile, the probability of misjudgment is also reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of image feature analysis, and in particular to a method and system for determining obstacles at a construction site based on image processing. Background Art

[0002] A construction site refers to the construction site where industrial and civil projects, such as housing construction, civil engineering, equipment installation, pipeline laying and other construction activities are carried out, the construction site that has been approved for occupation and the place where humans carry out safe production, civilized work and construction, including all areas on land, sea and air where construction work can be carried out.

[0003] A small amount of construction site image recognition can be performed with the naked eye. However, when a large number of construction site image recognitions need to be performed, if they are still performed with the naked eye, the workload of the recognition personnel will increase. Moreover, if the recognition personnel perform construction site image recognition for a long time, their concentration will decrease, and misjudgment may occur. Summary of the Invention

[0004] In order to solve the above technical problems, a method and system for determining construction site obstacles based on image processing are provided. This technical solution solves the problem proposed in the above background technology that a small amount of construction site image recognition can be identified and judged by the naked eye. When a large number of construction site image recognitions need to be identified and judged, if they are still identified and judged by the naked eye, the workload of the recognition personnel will be increased. Moreover, if the recognition personnel perform construction site image recognition for a long time, their concentration will decrease, and misjudgment may occur.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is: A method for determining obstacles at a construction site based on image processing, comprising: Read and process data from the database system to obtain a set of images of obstacles at the construction site; Preprocessing the construction site obstacle image set to obtain a second obstacle image set; Performing secondary image processing on the second obstacle image set according to parameters within the convolutional neural network model to obtain a third obstacle image set; Analyzing and processing the third obstacle image set to determine filter parameters of a convolutional layer of a convolutional neural network model; Inputting each image in the third obstacle image set into the convolutional neural network model for analysis and processing in sequence to obtain the type of obstacle to be verified; The obstacle type to be verified is verified through the third obstacle image set, and it is determined whether the convolutional neural network model needs to be retrained.

[0006] Preferably, the preprocessing of the construction site obstacle image set to obtain the second obstacle image set specifically includes the following steps: Using the fast Fourier transform algorithm, each image in the construction site obstacle image set is subjected to frequency domain conversion processing to obtain the spectrum information of each image in the construction site obstacle image set; Analyzing and processing frequency spectrum information of each image in the construction site obstacle image set to obtain a first obstacle image set; Performing image transformation processing on each image in the first obstacle image set to obtain a second obstacle image set; wherein the image transformation processing includes image rotation processing and image color transformation processing.

[0007] Preferably, analyzing and processing the spectrum information of each image in the construction site obstacle image set to obtain the first obstacle image set specifically comprises the following steps: By selecting a bandpass filter, the spectrum information of each image in the construction site obstacle image set is denoised to obtain the spectrum information of the obstacle image after the first preprocessing; Using the wavelet transform subband enhancement algorithm, the spectrum information of the obstacle image after the first preprocessing is enhanced to obtain the spectrum information of the obstacle image after the second preprocessing; The frequency spectrum information of the obstacle image after the second preprocessing is converted into the time domain by using an inverse Fourier transform algorithm to obtain a first obstacle image set.

[0008] Preferably, performing secondary image processing on the second obstacle image set according to the parameters within the convolutional neural network model to obtain the third obstacle image set specifically comprises the following steps: performing size data reading processing on each image in the second obstacle image set to determine the size data of each image in the second obstacle image set; Perform size data reading and processing on the input layer of the convolutional neural network model to determine the size data of the input image; Calculating and processing the size data of each image in the second obstacle image set and the size data of the input image to determine an image ratio that meets the input standard; The size of each image in the second obstacle image set is scaled according to an image ratio that meets the input standard to obtain a third obstacle image set.

[0009] Preferably, analyzing and processing the third obstacle image set to determine the filter parameters of the convolutional layer of the convolutional neural network model specifically includes the following steps: Analyzing and processing the channels of each image in the third obstacle image set to determine the number of channels of the obstacle image; Designing the filter of the convolutional layer of the convolutional neural network model according to the number of channels of the obstacle image, and determining the filter size information of the convolutional layer of the convolutional neural network model; The filter size information of the convolutional layer of the convolutional neural network model and the size information of the input image are analyzed and processed to determine the filter step size information of the convolutional layer of the convolutional neural network model.

[0010] Preferably, the step of sequentially inputting each image in the third obstacle image set into a convolutional neural network model for analysis and processing to obtain the obstacle type to be verified specifically comprises the following steps: performing convolution calculation processing on each region of each image in the third obstacle image set according to the filter of the convolution layer of the convolutional neural network model to obtain a feature value of each obstacle image; Perform image construction on the feature values ​​of each obstacle image according to the convolutional neural network model to obtain a first feature map; Activate the elements in the first feature map according to the ReLU activation function in the convolutional neural network model to obtain a second feature map; The second feature map is analyzed and processed according to the convolutional neural network model to obtain the type of obstacle to be verified.

[0011] Preferably, analyzing and processing the second feature map according to the convolutional neural network model to obtain the type of obstacle to be verified specifically includes the following steps: Performing size data reading processing on the second feature map according to the convolutional neural network model to determine the size data of the second feature map; Analyzing and processing the size data of the second feature map according to the convolutional neural network model to determine a pooling step size of a pooling layer of the convolutional neural network model; Performing feature selection processing on the second feature map according to the maximum pooling algorithm and the pooling step size of the pooling layer of the convolutional neural network model to determine a third feature map; The third feature map is analyzed and processed according to the convolutional neural network model to obtain the type of obstacle to be verified.

[0012] Preferably, analyzing and processing the third feature map according to the convolutional neural network model to obtain the type of obstacle to be verified specifically includes the following steps: Flatten the third feature map according to the convolutional neural network model to obtain one-dimensional feature data of the image; The one-dimensional feature data of the image is calculated and processed according to the matrix multiplication of the fully connected layer of the convolutional neural network model to obtain the one-dimensional feature vector of the image; According to the Softmax function in the convolutional neural network model, the one-dimensional feature vector of the image is classified and processed to determine the type of obstacle to be verified.

[0013] Preferably, verifying the obstacle type to be verified by using the third obstacle image set and selecting whether to retrain the convolutional neural network model specifically comprises the following steps: performing matching processing on the third obstacle image set according to the type of the obstacle to be verified; If the third obstacle image set contains an image associated with the type of obstacle to be verified, the convolutional neural network model can accurately identify the type of obstacle; If there is no image associated with the obstacle type to be verified in the third obstacle image set, the parameters inside the convolutional neural network model are changed, the convolutional neural network model is retrained, and the obstacle type to be verified is obtained again for verification.

[0014] Furthermore, a construction site obstacle determination system based on image processing is proposed, which is used to implement the above-mentioned construction site obstacle determination method based on image processing, including: An intelligent analysis terminal, which designs a convolutional neural network model by analyzing a collection of images of construction site obstacles. The intelligent analysis terminal is used to control data transmission and information exchange between various modules; A database system, the database system being used to store a set of images of construction site obstacles; An image preprocessing module, which performs denoising and enhancement on the construction site obstacle image set to obtain a second obstacle image set; an image resizing module, configured to resize each image in the second obstacle image set to determine a third obstacle image set; a convolutional neural network model, the convolutional neural network model being used to perform obstacle type analysis on each image in the third obstacle image set to obtain the obstacle type to be verified; A type verification module verifies the type of the obstacle to be verified based on the third obstacle image set and selects whether the convolutional neural network model needs to be retrained.

[0015] Compared with the existing technology, the present invention provides a method and system for determining construction site obstacles based on image processing, which has the following beneficial effects: The present invention first preprocesses a set of construction site obstacle images to remove noise from the images and also performs feature enhancement on the images to improve the recognition of image features. Then, the first obstacle image set is rotated and color-transformed to expand the first obstacle image set. The third obstacle image set is then feature-recognized using a convolutional neural network model to determine the type of obstacles to be verified. Finally, the type of obstacles to be verified is verified using the third obstacle image set to determine the accuracy of the convolutional neural network model. If the accuracy of the convolutional neural network model meets the standard, it is put into use. Using the convolutional neural network model to recognize images can reduce the workload of recognition personnel and reduce the probability of misjudgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the flow of steps S100-S600 in a method for determining obstacles at a construction site based on image processing proposed by the present invention; Figure 2 This is a structural block diagram of a construction site obstacle determination system based on image processing proposed by the present invention. DETAILED DESCRIPTION

[0017] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0018] Reference Figure 1 As shown, a method for determining obstacles at a construction site based on image processing includes: S100, reading and processing data from a database system to obtain a set of images of obstacles at the construction site; S200: Preprocessing the construction site obstacle image set to obtain a second obstacle image set; S300: Perform secondary image processing on the second obstacle image set according to parameters within the convolutional neural network model to obtain a third obstacle image set; S400: Analyze and process the third obstacle image set to determine filter parameters of a convolutional layer of a convolutional neural network model; S500: Input each image in the third obstacle image set into the convolutional neural network model for analysis and processing to obtain the type of obstacle to be verified; S600: Verify the obstacle type to be verified using the third obstacle image set, and select whether to retrain the convolutional neural network model; It will be understood by those skilled in the art that the feature recognition of a small number of construction site obstacle images can be identified and judged by the naked eye. When a large number of construction site obstacle images need to be feature recognized, if they are recognized by the naked eye, it will not only increase the workload of the recognition personnel, but may also lead to misjudgment. Therefore, when the recognition personnel perform feature recognition of construction site obstacle images for a long time, it will cause a decrease in concentration, which indirectly increases the probability of misjudgment. Therefore, by constructing a convolutional neural network model for image recognition, it can effectively reduce the workload of the recognition personnel and reduce the probability of misjudgment. When the staff misjudges the construction site obstacle image, it may indirectly cause a certain degree of harm to the worker's body. For example, there is an obstacle in a certain area of ​​the construction site. Due to the staff's misjudgment, a warning sign is not erected near the obstacle, and the worker may be injured by the obstacle when passing by. Example

[0019] S200, pre-processing the construction site obstacle image set to obtain a second obstacle image set specifically includes the following steps: S201. Perform frequency domain conversion processing on each image in the construction site obstacle image set using a fast Fourier transform algorithm to obtain frequency spectrum information of each image in the construction site obstacle image set. S202: Analyze and process the frequency spectrum information of each image in the construction site obstacle image set to obtain a first obstacle image set; S203: performing image transformation processing on each image in the first obstacle image set to obtain a second obstacle image set; wherein the image transformation processing includes image rotation processing and image color transformation processing; In this embodiment, the construction site obstacle image set is preprocessed to remove interference factors in the image and improve the recognition accuracy of the convolutional neural network model. In addition, each image in the first obstacle image set is rotated and color converted to enable the convolutional neural network model to recognize the same image in different environments, thereby indirectly improving the accuracy of the convolutional neural network model and avoiding multiple training of the convolutional neural network model. S202, analyzing and processing the frequency spectrum information of each image in the construction site obstacle image set to obtain the first obstacle image set, specifically includes the following steps: S2021. Perform denoising on the spectrum information of each image in the set of construction site obstacle images by selecting a bandpass filter to obtain spectrum information of the obstacle image after the first preprocessing. S2022. Perform spectrum enhancement processing on the spectrum information of the obstacle image after the first preprocessing by using a wavelet transform subband enhancement algorithm to obtain spectrum information of the obstacle image after the second preprocessing; S2023. Performing time-domain conversion processing on the spectrum information of the obstacle image after the second preprocessing using an inverse Fourier transform algorithm to obtain a first obstacle image set; In this embodiment, the images in the construction site obstacle image set may contain noise, and the noise in the images may affect the accuracy of the convolutional neural network model. Therefore, the construction site obstacle image set is denoised. At the same time, in order to make the convolutional neural network model recognize the features of the image more quickly, the features of each image in the construction site obstacle image set are enhanced. Example

[0020] S300, performing secondary image processing on the second obstacle image set according to the internal parameters of the convolutional neural network model to obtain a third obstacle image set specifically includes the following steps: S301: performing size data reading processing on each image in the second obstacle image set to determine the size data of each image in the second obstacle image set; S302, performing size data reading processing on the input layer of the convolutional neural network model to determine the size data of the input image; S303: Calculate and process the size data of each image in the second obstacle image set and the size data of the input image to determine an image ratio that meets the input standard; S304: scaling each image in the second obstacle image set according to an image ratio that meets the input standard to obtain a third obstacle image set; In this embodiment, the input layer of the convolutional neural network model can only receive images of a fixed size. Therefore, the input layer of the convolutional neural network model is analyzed to determine the size information of the input data. Then, the image size data in the second obstacle image set and the size data of the input data are calculated and processed to determine the image ratio that meets the input standard. Each image in the second obstacle image set is resized according to the image ratio that meets the input standard to obtain an image that is compatible with the input layer of the convolutional neural network model. Example

[0021] S400, analyzing and processing the third obstacle image set to determine the filter parameters of the convolutional layer of the convolutional neural network model specifically includes the following steps: S401: Analyze and process the channels of each image in the third obstacle image set to determine the number of channels of the obstacle image; S402: Designing a filter of a convolutional layer of the convolutional neural network model according to the number of channels of the obstacle image, and determining filter size information of the convolutional layer of the convolutional neural network model; S403: Analyze and process the filter size information of the convolutional layer of the convolutional neural network model and the size information of the input image to determine the filter step size information of the convolutional layer of the convolutional neural network model; In this embodiment, the size and depth of the filter are related to the channels of the image. For example, an RGB image has three channels: red, green, and blue. Therefore, the size information of the filter should be designed to be 3*3*3. Therefore, it is necessary to analyze the number of image channels to determine the size information of the filter, and the step size information of the filter determines the size of the first feature map. In order to keep the size of the feature map unchanged, either the feature map is padded or the step size of the filter is changed. However, the complexity of filling the feature map is greater than the step size design of the filter. Therefore, the size of the feature map is ensured to be unchanged by designing the step size of the filter. Example

[0022] S500: Inputting each image in the third obstacle image set into the convolutional neural network model for analysis and processing, and obtaining the obstacle type to be verified specifically includes the following steps: S501: Perform convolution calculation processing on each region of each image in the third obstacle image set according to the filter of the convolution layer of the convolutional neural network model to obtain a feature value of each obstacle image; S502: construct an image based on the feature values ​​of each obstacle image according to the convolutional neural network model to obtain a first feature map; S503: Activate the elements in the first feature map according to the ReLU activation function in the convolutional neural network model to obtain a second feature map; S504: Analyze and process the second feature map according to the convolutional neural network model to obtain the type of obstacle to be verified; In this embodiment, in order to reduce the subsequent amount of calculation, the first feature map is activated by the ReLU activation function in the convolutional neural network model, the sparsity of the first feature map is improved, and some elements in the first feature map are changed to zero, thereby reducing the amount of calculation for subsequent image feature recognition.

[0023] S504, analyzing and processing the second feature map according to the convolutional neural network model to obtain the type of obstacle to be verified, specifically includes the following steps: S5041. Perform size data reading processing on the second feature map according to the convolutional neural network model to determine the size data of the second feature map; S5042. Analyze and process the size data of the second feature map according to the convolutional neural network model to determine a pooling step size of a pooling layer of the convolutional neural network model; S5043. Perform feature selection processing on the second feature map according to the maximum pooling algorithm and the pooling step size of the pooling layer of the convolutional neural network model to determine a third feature map; S5044. Analyze and process the third feature map according to the convolutional neural network model to obtain the type of obstacle to be verified; In this embodiment, the more obvious the features in the obstacle image are, the larger the pixel value of the obstacle image is. Therefore, the second feature map is pooled by the pooling layer, and the second feature map is pooled by the maximum pooling algorithm. Figure 1 The maximum value within a certain range is filtered out, which is the feature of the obstacle image. However, when some features of the obstacle image change drastically within a certain range, more and larger feature values ​​will appear. In order to make feature recognition more accurate, the size data of the second feature map is analyzed and the pooling step size of the pooling layer is reduced.

[0024] S5044, analyzing and processing the third feature map according to the convolutional neural network model to obtain the type of obstacle to be verified, specifically includes the following steps: S50441. Perform image dimension flattening processing on the third feature map according to the convolutional neural network model to obtain one-dimensional feature data of the image; S50442. Calculate and process the one-dimensional feature data of the image according to the matrix multiplication of the fully connected layer of the convolutional neural network model to obtain a one-dimensional feature vector of the image; S50443. Perform vector classification processing on the one-dimensional feature vector of the image based on the Softmax function in the convolutional neural network model to determine the type of obstacle to be verified; In this embodiment, since the output of the convolutional or pooling layer is a two-dimensional feature map, and the fully connected layer requires a one-dimensional input, the two-dimensional feature map must first be flattened to convert the multidimensional input data into one-dimensional data. The flattened feature map is then input into the fully connected layer and converted into a fixed-length one-dimensional feature vector through matrix multiplication. This process achieves the mapping from high-dimensional feature space to low-dimensional feature space while retaining the most useful information. Finally, the fully connected layer performs classification or regression operations on the one-dimensional feature vector using an activation function (such as the Softmax function). In classification tasks, the Softmax function converts the feature vector into a probability distribution, representing the likelihood of each category; the likelihood of each category is then determined and a predicted value is output. Example

[0025] S600: Verifying the obstacle type to be verified using the third obstacle image set and selecting whether to retrain the convolutional neural network model specifically includes the following steps: S601: performing matching processing on the third obstacle image set according to the type of the obstacle to be verified; S602: If the third obstacle image set includes an image associated with the obstacle type to be verified, the convolutional neural network model can accurately identify the obstacle type; S603: If the third obstacle image set does not contain an image associated with the obstacle type to be verified, modify the parameters within the convolutional neural network model, retrain the convolutional neural network model, and obtain the obstacle type to be verified again for verification; In this embodiment, the convolutional neural network model needs to be analyzed for its judgment accuracy before it is put into use. If the accuracy of the convolutional neural network model is not enough, it needs to be retrained to avoid increasing the misjudgment rate of image features.

[0026] Reference Figure 2 As shown, a construction site obstacle determination system based on image processing is used to implement the above-mentioned construction site obstacle determination method based on image processing, including: An intelligent analysis terminal, which designs a convolutional neural network model by analyzing a collection of images of construction site obstacles. The intelligent analysis terminal is used to control data transmission and information exchange between various modules; A database system, the database system being used to store a set of images of construction site obstacles; An image preprocessing module, which performs denoising and enhancement on the construction site obstacle image set to obtain a second obstacle image set; an image resizing module, configured to resize each image in the second obstacle image set to determine a third obstacle image set; a convolutional neural network model, the convolutional neural network model being used to perform obstacle type analysis on each image in the third obstacle image set to obtain the obstacle type to be verified; A type verification module verifies the type of the obstacle to be verified based on the third obstacle image set and selects whether the convolutional neural network model needs to be retrained.

[0027] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for determining obstacles at a construction site based on image processing, characterized in that: include: Read and process data from the database system to obtain a set of images of obstacles at the construction site; Preprocessing the construction site obstacle image set to obtain a second obstacle image set; Performing secondary image processing on the second obstacle image set according to parameters within the convolutional neural network model to obtain a third obstacle image set; Analyzing and processing the third obstacle image set to determine filter parameters of a convolutional layer of a convolutional neural network model; Inputting each image in the third obstacle image set into the convolutional neural network model for analysis and processing in sequence to obtain the type of obstacle to be verified; The obstacle type to be verified is verified through the third obstacle image set, and it is determined whether the convolutional neural network model needs to be retrained.

2. The method for determining obstacles at a construction site based on image processing according to claim 1, characterized in that: Preprocessing the construction site obstacle image set to obtain the second obstacle image set specifically includes the following steps: Using the fast Fourier transform algorithm, each image in the construction site obstacle image set is subjected to frequency domain conversion processing to obtain the spectrum information of each image in the construction site obstacle image set; Analyzing and processing frequency spectrum information of each image in the construction site obstacle image set to obtain a first obstacle image set; Performing image transformation processing on each image in the first obstacle image set to obtain a second obstacle image set; wherein the image transformation processing includes image rotation processing and image color transformation processing.

3. The method for determining obstacles at a construction site based on image processing according to claim 2, characterized in that: Analyzing and processing the spectrum information of each image in the construction site obstacle image set to obtain the first obstacle image set specifically includes the following steps: By selecting a bandpass filter, the spectrum information of each image in the construction site obstacle image set is denoised to obtain the spectrum information of the obstacle image after the first preprocessing; Using the wavelet transform subband enhancement algorithm, the spectrum information of the obstacle image after the first preprocessing is enhanced to obtain the spectrum information of the obstacle image after the second preprocessing; The frequency spectrum information of the obstacle image after the second preprocessing is converted into the time domain by using an inverse Fourier transform algorithm to obtain a first obstacle image set.

4. The method for determining obstacles at a construction site based on image processing according to claim 1, characterized in that: The performing secondary image processing on the second obstacle image set according to the parameters within the convolutional neural network model to obtain the third obstacle image set specifically includes the following steps: performing size data reading processing on each image in the second obstacle image set to determine the size data of each image in the second obstacle image set; Perform size data reading and processing on the input layer of the convolutional neural network model to determine the size data of the input image; Calculating and processing the size data of each image in the second obstacle image set and the size data of the input image to determine an image ratio that meets the input standard; The size of each image in the second obstacle image set is scaled according to an image ratio that meets the input standard to obtain a third obstacle image set.

5. The method for determining obstacles at a construction site based on image processing according to claim 1, characterized in that: The analyzing and processing the third obstacle image set to determine the filter parameters of the convolutional layer of the convolutional neural network model specifically includes the following steps: Analyzing and processing the channels of each image in the third obstacle image set to determine the number of channels of the obstacle image; Designing the filter of the convolutional layer of the convolutional neural network model according to the number of channels of the obstacle image, and determining the filter size information of the convolutional layer of the convolutional neural network model; The filter size information of the convolutional layer of the convolutional neural network model and the size information of the input image are analyzed and processed to determine the filter step size information of the convolutional layer of the convolutional neural network model.

6. The method for determining obstacles at a construction site based on image processing according to claim 1, characterized in that: Inputting each image in the third obstacle image set into the convolutional neural network model for analysis and processing to obtain the obstacle type to be verified specifically includes the following steps: performing convolution calculation processing on each region of each image in the third obstacle image set according to the filter of the convolution layer of the convolutional neural network model to obtain a feature value of each obstacle image; Perform image construction on the feature values ​​of each obstacle image according to the convolutional neural network model to obtain a first feature map; Activate the elements in the first feature map according to the ReLU activation function in the convolutional neural network model to obtain a second feature map; The second feature map is analyzed and processed according to the convolutional neural network model to obtain the type of obstacle to be verified.

7. The method for determining obstacles at a construction site based on image processing according to claim 6, characterized in that: Analyzing and processing the second feature map according to the convolutional neural network model to obtain the type of obstacle to be verified specifically includes the following steps: Performing size data reading processing on the second feature map according to the convolutional neural network model to determine the size data of the second feature map; Analyzing and processing the size data of the second feature map according to the convolutional neural network model to determine a pooling step size of a pooling layer of the convolutional neural network model; Performing feature selection processing on the second feature map according to the maximum pooling algorithm and the pooling step size of the pooling layer of the convolutional neural network model to determine a third feature map; The third feature map is analyzed and processed according to the convolutional neural network model to obtain the type of obstacle to be verified.

8. The method for determining obstacles at a construction site based on image processing according to claim 1, characterized in that: Analyzing and processing the third feature map according to the convolutional neural network model to obtain the type of obstacle to be verified specifically includes the following steps: Flatten the third feature map according to the convolutional neural network model to obtain one-dimensional feature data of the image; The one-dimensional feature data of the image is calculated and processed according to the matrix multiplication of the fully connected layer of the convolutional neural network model to obtain the one-dimensional feature vector of the image; According to the Softmax function in the convolutional neural network model, the one-dimensional feature vector of the image is classified and processed to determine the type of obstacle to be verified.

9. The method for determining obstacles at a construction site based on image processing according to claim 1, characterized in that: Verifying the obstacle type to be verified by using the third obstacle image set and selecting whether to retrain the convolutional neural network model specifically includes the following steps: performing matching processing on the third obstacle image set according to the type of the obstacle to be verified; If the third obstacle image set contains an image associated with the type of obstacle to be verified, the convolutional neural network model can accurately identify the type of obstacle; If there is no image associated with the obstacle type to be verified in the third obstacle image set, the parameters inside the convolutional neural network model are changed, the convolutional neural network model is retrained, and the obstacle type to be verified is obtained again for verification.

10. A construction site obstacle determination system based on image processing, used to implement the construction site obstacle determination method based on image processing according to any one of claims 1 to 9, characterized in that: include: An intelligent analysis terminal, which designs a convolutional neural network model by analyzing a collection of images of construction site obstacles. The intelligent analysis terminal is used to control data transmission and information exchange between various modules; A database system, the database system being used to store a set of images of construction site obstacles; An image preprocessing module, which performs denoising and enhancement on the construction site obstacle image set to obtain a second obstacle image set; an image resizing module, configured to resize each image in the second obstacle image set to determine a third obstacle image set; a convolutional neural network model, the convolutional neural network model being used to perform obstacle type analysis on each image in the third obstacle image set to obtain the obstacle type to be verified; A type verification module verifies the type of the obstacle to be verified based on the third obstacle image set and selects whether the convolutional neural network model needs to be retrained.

Images