Intelligent safety helmet construction environment detection method and system based on image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to an intelligent safety helmet construction environment detection method and system based on image recognition, and the method comprises the steps: obtaining the image data of a construction environment; preprocessing the acquired image data, including denoising the construction image and enhancing the denoised image data; the preprocessed image is used as input to construct a prediction model, and an attention mechanism is introduced for the prediction model; the method comprises the steps of constructing a prediction model based on a convolutional neural network, a recurrent neural network and a generative adversarial network, performing model training according to the constructed prediction model, performing feature extraction based on the trained prediction model, and performing judgment of a danger threshold according to output image features, and can fully utilize the advantages of different networks by constructing the prediction model based on the convolutional neural network, the recurrent neural network and the generative adversarial network. Therefore, the judgment on dangerous conditions is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to an intelligent safety helmet construction environment detection method and system based on image recognition. Background Art

[0002] In various construction scenarios, hard hats, as essential basic safety equipment for construction workers, play a vital role. They can effectively protect the heads of construction workers from being hit by objects and other injuries when accidents occur. However, the functions of traditional hard hats are relatively simple, and they only provide physical protection. They lack the ability to actively detect and warn of potential dangers in the construction environment. Traditional hard hats mainly focus on physical protection of the head and cannot monitor the construction environment in real time. When wearing traditional hard hats, construction workers can only rely on their own observations and experience to judge the safety of the surrounding environment, which can easily lead to misjudgment or omission of potential dangers in complex construction environments.

[0003] In addition, traditional hard hats lack integration with modern technology and cannot achieve automated environmental detection and early warning. During the construction process, construction workers may ignore changes in the surrounding environment because they are focused on their work, resulting in failure to respond in time when danger occurs. Some construction sites use sensors for environmental detection, such as temperature sensors, humidity sensors, gas sensors, etc. Although sensors can monitor specific physical parameters in real time, their detection range is limited and they can often only detect a single type of dangerous factor. They cannot make accurate judgments based on the specific work positions of construction workers. Moreover, sensors have weak object recognition and danger judgment capabilities in complex construction environments and cannot comprehensively evaluate the safety status of the construction environment. At this stage, a construction environment detection method and system for intelligent hard hats based on image recognition is needed. Summary of the invention

[0004] In order to solve the problem that traditional hard hats have weak object recognition and hazard judgment capabilities in complex construction environments, the present invention provides an intelligent hard hat construction environment detection method and system based on image recognition. The present invention combines image recognition technology with hard hats to accurately identify dangerous factors in the construction environment and improve the accuracy and efficiency of detection.

[0005] In the first aspect, the present invention provides a method for detecting a construction environment of an intelligent safety helmet based on image recognition, which adopts the following technical solution: An intelligent safety helmet construction environment detection method based on image recognition, comprising: Acquire image data of the construction environment; Preprocessing the acquired image data, including denoising the construction image and enhancing the denoised image data; Use the preprocessed images as input to build prediction models, including prediction models based on convolutional neural networks, recurrent neural networks, and generative adversarial networks, and introduce attention mechanisms for the prediction models; Performing model training based on the constructed prediction model, including adjusting the parameters of the prediction model using an adaptive optimization algorithm; Perform feature extraction based on the trained prediction model, including performing convolution operations with the input image through the convolution kernel to extract local features of the image; The danger threshold is determined based on the output image features.

[0006] Furthermore, the construction image is denoised and the denoised image data is enhanced, including using a median filter to replace the grayscale value of each pixel in the image data with the median of the grayscale values ​​in the neighborhood of the pixel, then calculating the grayscale histogram of the original image through histogram equalization, calculating the cumulative distribution function based on the grayscale histogram, and mapping the grayscale values ​​of the original image through the cumulative distribution function to obtain an enhanced image.

[0007] Furthermore, the construction of a prediction model based on a convolutional neural network, a recurrent neural network and a generative adversarial network includes constructing a convolutional neural network consisting of multiple convolutional layers, activation function layers, pooling layers and fully connected layers, wherein the convolutional layer is composed of multiple convolutional kernels of different sizes, and the input image data is passed through multiple convolutional layers to obtain a set of feature maps. The different feature maps output by the convolutional layer are input to the activation function layer for nonlinear transformation, and then enter the pooling layer for downsampling operation, and finally the fully connected layer performs feature integration.

[0008] Furthermore, the construction of a prediction model based on a convolutional neural network, a recurrent neural network and a generative adversarial network also includes the recurrent neural network adopting a long short-term memory network, the input gate of the recurrent neural network is used to receive feature information output by a fully connected layer, the output gate of the recurrent neural network transmits the output image data information to the generator of the generative adversarial network, and the realistic construction environment image sequence generated by the adversarial network is used as additional training data and input into the recurrent neural network for training.

[0009] Furthermore, the attention mechanism is introduced into the prediction model, including obtaining two feature maps through two convolutional layers for the input feature map respectively, and then performing element-by-element multiplication and sigmoid activation on the two feature maps to obtain a spatial attention map, and finally, performing element-by-element multiplication of the input feature map and the spatial attention map to obtain an enhanced feature map.

[0010] Furthermore, the use of the adaptive optimization algorithm to adjust the parameters of the model includes calculating the exponential moving average of the square of the model gradient according to each parameter in the prediction model, and updating the model parameters using the exponential moving average. The calculation formula of the exponential moving average is: , in, It is expressed as the exponential moving average of the square of the gradient at time t, It is expressed as a decay rate, which is a value between 0 and 1. It is expressed as the exponential moving average of the square of the gradient at the previous moment, It is expressed as the gradient at the current moment, that is, the derivative of the loss function with respect to the current parameters.

[0011] Furthermore, the judgment of the danger threshold based on the output image features includes dividing the image feature results output by the prediction model into dangerous equipment features, personnel behavior features and environmental condition features, constructing a danger assessment model by assigning weights to different features, and using the output structure of the danger assessment model to judge the danger threshold.

[0012] In the second aspect, an intelligent safety helmet construction environment detection system based on image recognition includes: The data acquisition module is configured to: acquire image data of the construction environment; The preprocessing module is configured to: preprocess the acquired image data, including denoising the construction image, and performing image enhancement on the denoised image data; The conversion module is configured to: use the preprocessed image as input to build a prediction model, including building a prediction model based on a convolutional neural network, a recurrent neural network, and a generative adversarial network, and introduce an attention mechanism for the prediction model; The model module is configured to: perform model training according to the constructed prediction model, including adjusting the parameters of the prediction model by using an adaptive optimization algorithm; The feature extraction module is configured to: perform feature extraction based on the trained prediction model, including performing a convolution operation with the input image through a convolution kernel to extract local features of the image; The transformation module is configured to: determine the danger threshold according to the output image features.

[0013] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a method for detecting a construction environment of an intelligent safety helmet based on image recognition.

[0014] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor as a method for detecting a construction environment of an intelligent safety helmet based on image recognition.

[0015] In summary, the present invention has the following beneficial technical effects: 1. The present invention can fully utilize the advantages of different networks by constructing a prediction model based on convolutional neural networks, recurrent neural networks and generative adversarial networks. The convolutional neural network extracts local features of the image and captures information such as the shape and texture of the object; the recurrent neural network processes the image sequence, records the dynamic changes during the construction process, and better identifies the development trend of potential dangers.

[0016] 2. The present invention further enhances the model's attention to key areas by introducing an attention mechanism, thereby improving the accuracy of feature extraction. For the input feature map, two feature maps are obtained through two convolutional layers and processed to obtain a spatial attention map, which is then multiplied element by element with the input feature map, so that the model pays more attention to important areas, thereby improving the accuracy of identifying dangerous situations.

[0017] 3. The present invention uses an adaptive optimization algorithm to adjust the parameters of the prediction model. When the gradient is large, the learning rate is automatically adjusted to a larger value so as to approach the optimal solution more quickly. When the gradient is small, the learning rate is reduced to avoid oscillation near the optimal solution. This allows the model to be efficiently trained in different construction environments and tasks, thereby improving the generalization ability of the model.

[0018] 4. The present invention determines the danger threshold based on the output image features, divides the image feature results into dangerous equipment features, personnel behavior features and environmental condition features, and constructs a danger assessment model by assigning weights to different features. This multi-factor comprehensive judgment method can quickly and accurately assess the danger of the construction environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the overall process of an intelligent safety helmet construction environment detection method based on image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0021] Example 1 Reference Figure 1 , a method for detecting a construction environment of an intelligent safety helmet based on image recognition in this embodiment includes: Acquire image data of the construction environment; Preprocessing the acquired image data, including denoising the construction image and enhancing the denoised image data; Use the preprocessed images as input to build prediction models, including prediction models based on convolutional neural networks, recurrent neural networks, and generative adversarial networks, and introduce attention mechanisms for the prediction models; Performing model training based on the constructed prediction model, including adjusting the parameters of the prediction model using an adaptive optimization algorithm; Perform feature extraction based on the trained prediction model, including performing convolution operations with the input image through the convolution kernel to extract local features of the image; The danger threshold is determined based on the output image features.

[0022] Specifically, a method for detecting a construction environment of an intelligent safety helmet based on image recognition comprises the following steps: like Figure 1 As shown, S1, obtaining image data of the construction environment; Choose a high-resolution camera to install on the smart helmet to ensure that it can clearly capture images of the construction environment. Before installing the camera, it is necessary to analyze the overall layout of the construction site to understand the location and importance of different construction stages and different work areas. For example, for key areas such as high-altitude work areas, large equipment operation areas, and hazardous material storage areas, it should be ensured that the camera can be directly aimed at or have a good viewing angle coverage.

[0023] S2, preprocessing the acquired image data, including denoising the construction image, and performing image enhancement on the denoised image data; The median filter is used to denoise the construction image. For a construction image, the gray value of its pixel is ,in Represents the coordinates of the pixel point and determines the size of a neighborhood, for example or For each pixel in the image , taking the pixel as the center, select all the pixels in its neighborhood, sort the grayscale values ​​of the pixels in the neighborhood, and take the middle value as the new grayscale value of the central pixel. The calculation formula is: , in is the original image at coordinates The gray value at is the filtered image at coordinates The gray value at , W is the image window, Represents the median operation.

[0024] After filtering, the denoised image is enhanced by histogram equalization.

[0025] First, calculate the grayscale histogram of the original image , where i represents the gray level, Represents the number of pixels with gray value i in the image, and then calculates the cumulative distribution function based on the gray histogram , , where j represents the gray level, Represents the cumulative number of pixels in the image whose grayscale value is less than or equal to i. Finally, the grayscale value of the original image is mapped by the cumulative distribution function to obtain the enhanced image. Suppose the grayscale value of the pixel in the original image is , the enhanced gray value is: , Where N is the total number of pixels in the image and L is the total number of gray levels.

[0026] S3. Use the preprocessed image as input to build a prediction model, including building a prediction model based on a convolutional neural network, a recurrent neural network, and a generative adversarial network, and introduce an attention mechanism for the prediction model; The convolutional neural network (CNN) consists of an input layer, a convolution layer, an activation function layer, a pooling layer, and a fully connected layer. The input layer receives the preprocessed construction environment image as input. The image is represented in the form of a matrix, where each element represents the color value or grayscale value of a pixel. The convolution layer uses three convolution kernels of different scales, namely , and For the input image After the convolution operation, feature maps of different scales are obtained. Convolution kernel, output feature map ,in, and is the height and width of the output feature map, is the number of channels, and the convolution operation is expressed as: , in, Represented as a convolution operation, Expressed as The convolution kernel, It is expressed as a bias term. The convolution operation is to slide the convolution kernel pixel by pixel on the input image and perform element-level multiplication and summation operations at each position. After the convolution layer outputs the feature map, it enters the activation function layer, where the activation function adopts the ReLU activation function. The ReLU function remains unchanged when the input is a positive number and outputs zero when the input is a negative number, thereby introducing nonlinear characteristics. After that, the pooling layer is used to downsample the feature map to reduce the resolution of the feature map, reduce the amount of calculation and the number of parameters. The pooling operation adopts the maximum pooling operation, by selecting the maximum value of each area in the feature map as the output. Finally, the fully connected layer performs feature integration to integrate the features extracted by the previous layer and output the final classification or prediction results. Each neuron in the fully connected layer is connected to all neurons in the previous layer, and linear transformation and nonlinear transformation of the activation function are performed through the weight matrix. The last fully connected layer is usually used as the output layer to output the corresponding results according to the task.

[0027] The recurrent neural network uses the long short-term memory network LSTM. First, the construction environment image is input into the CNN for feature extraction. After being processed by multiple convolutional layers, activation function layers, and pooling layers, CNN obtains a set of feature maps. These feature maps have a high dimension and can be converted into one-dimensional vectors through flattening operations. The flattened feature vector is used as the input of a time step of the LSTM input sequence. The feature map enters the input gate, forget gate, cell state, and output gate of the LSTM in turn. The input gate performs a nonlinear transformation on the current input and the hidden state of the previous moment to obtain a value between 0 and 1, which indicates the importance of the input information. The calculation formula is: , in, is the input at the current moment, is the hidden state at the previous moment, is the cell state at the previous moment, , and is the weight matrix, is the bias term, is a sigmoid function, and the forget gate determines which information will be forgotten. The forget gate performs a nonlinear transformation on the current input and the hidden state of the previous moment to obtain a value between 0 and 1, indicating which information in the cell state of the previous moment needs to be forgotten. Its calculation formula is the same as that of the input gate, and the cell state is used to store long-term memory. The cell state is controlled by the input gate and the forget gate to store new information and forget old information, thereby updating the cell state. The calculation formula is: , in, is the output of the forget gate, is the element-wise product, is the hyperbolic tangent function, is the bias term, and is a weight matrix, and finally the output gate is used to output the generated adversarial network GAN. GAN is mainly used to generate realistic construction environment image data, expand the training data set, and improve the generalization ability of the model. GAN consists of a generator and a discriminator. The goal of the generator is to generate fake images similar to real images to deceive the discriminator; the goal of the discriminator is to distinguish between real images and fake images generated by the generator.

[0028] When training GAN, the parameters of the generator and the discriminator are optimized alternately so that the generator gradually generates more realistic images. Random noise can be used as the input of the generator to generate fake images through a multi-layer neural network. The discriminator receives real images and fake images as input and outputs a probability value, which indicates the probability that the input image is a real image. In summary, the overall construction process of the prediction model is to input the real construction environment image into CNN for feature extraction. The feature vector sequence output by CNN is input into LSTM, which learns the temporal dependency of the image sequence. At the same time, the output of LSTM can be fed back to CNN for feature fusion optimization. On the other hand, the output of LSTM can also guide the generator of GAN to generate new image sequences. The image sequence generated by GAN can be input into CNN and LSTM together with the real image for training to form a closed-loop system.

[0029] After the model is built, the attention mechanism is introduced. First, the spatial attention map is calculated. For the input feature map , two feature maps are obtained through two convolutional layers and , then these two feature maps are element-wise multiplied and sigmoid activated to obtain the spatial attention map ,Right now Finally, the input feature map is multiplied element by element with the spatial attention map to obtain the enhanced feature map .

[0030] S4. Performing model training based on the constructed prediction model, including adjusting the parameters of the prediction model using an adaptive optimization algorithm; Set initial parameter values, including learning rate , attenuation rate (between 0 and 1), constant (used to prevent the denominator from being zero), etc. At the same time, initialize the exponential moving average of the squared gradient is a small positive number. For each parameter in the prediction model, the derivative of the loss function with respect to the parameter is calculated, that is, the gradient at the current moment. , the exponential moving average of the square of the model gradient is calculated according to each parameter in the prediction model, and the exponential moving average is used to update the model parameters. The calculation formula of the exponential moving average is: , in, It is expressed as the exponential moving average of the square of the gradient at time t, It is expressed as a decay rate, which is a value between 0 and 1. It is expressed as the exponential moving average of the square of the gradient at the previous moment, It is expressed as the gradient at the current moment, that is, the derivative of the loss function with respect to the current parameters.

[0031] , in, is the current parameter value, is the learning rate, is a constant used to prevent the denominator from being zero. For example, the current parameter , learning rate , the current gradient , the exponential moving average calculated in the previous step , then the updated parameters are: , New parameter values ​​can be obtained through calculation.

[0032] S5. performing feature extraction based on the trained prediction model, including performing a convolution operation on the input image through a convolution kernel to extract local features of the image; The construction image data obtained on site is uploaded to the prediction model. The convolution layer in the CNN performs convolution operation with the input image through the convolution kernel to extract the local features of the image. An input image of size The convolution kernel of the convolution operation is calculated as: , in, is the input image at coordinates The gray value at is the image after convolution at coordinates The gray value at is the convolution kernel at coordinates The weight of It is a bias term. The pooling layer downsamples the convolution image to reduce the image resolution, reduce the amount of calculation, and extract the global features of the image.

[0033] This embodiment adopts the maximum pooling. An input image of size The pooling window of the maximum pooling is calculated as: , in is the input image at coordinates The gray value at is the image after pooling at coordinates The gray value at is the pooling window.

[0034] S6. Determine the danger threshold based on the output image features.

[0035] According to the image feature results output by the prediction model, it is marked as dangerous equipment features, personnel behavior features and environmental condition features. Among them, dangerous equipment features include dangerous equipment on the construction site and the shape, color and logo of the equipment. Personnel behavior features include the distance between construction personnel and dangerous areas. Environmental condition features include whether there are obstacles and dangerous areas (such as deep pits, edges of aerial work platforms, etc.) in the current construction environment. Weights are assigned to different features to build a hazard assessment model. The weight assignment uses the analytic hierarchy process (AHP) to determine the weight of each feature. First, a judgment matrix is ​​constructed to compare the relative importance of different features. Then, the weight of each feature is determined by calculating the eigenvector of the judgment matrix, and the hazard assessment model is constructed using the linear weighted method. The expression of the hazard assessment model is: , in, represents the risk assessment value, , and Respectively represent the values ​​of dangerous equipment characteristics, personnel behavior characteristics and environmental conditions characteristics, , and They represent the weight of each feature respectively, and different thresholds are set according to different types of construction projects. For example, bridge construction usually involves high-altitude operations and large-scale machinery operations, which are more dangerous. The danger threshold can be set to 65. Road construction is mainly ground operations, which are relatively less dangerous. The danger threshold can be set to 75. The danger assessment value is compared with the threshold, and a danger warning is issued when the threshold is exceeded.

[0036] Example 2 The difference between this embodiment and embodiment 1 is that this embodiment provides an intelligent safety helmet construction environment detection system based on image recognition, including: The data acquisition module is configured to: acquire image data of the construction environment; The preprocessing module is configured to: preprocess the acquired image data, including denoising the construction image, and performing image enhancement on the denoised image data; The conversion module is configured to: use the preprocessed image as input to build a prediction model, including building a prediction model based on a convolutional neural network, a recurrent neural network, and a generative adversarial network, and introduce an attention mechanism for the prediction model; The model module is configured to: perform model training according to the constructed prediction model, including adjusting the parameters of the prediction model by using an adaptive optimization algorithm; The feature extraction module is configured to: perform feature extraction based on the trained prediction model, including performing a convolution operation with the input image through a convolution kernel to extract local features of the image; The transformation module is configured to: determine the danger threshold according to the output image features.

[0037] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a method for detecting a construction environment of an intelligent safety helmet based on image recognition.

[0038] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor, and the method for detecting a construction environment of an intelligent safety helmet based on image recognition is described.

[0039] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting construction environment of intelligent safety helmets based on image recognition, characterized in that: include: Acquire image data of the construction environment; Preprocessing the acquired image data, including denoising the construction image and enhancing the denoised image data; Use the preprocessed images as input to build prediction models, including prediction models based on convolutional neural networks, recurrent neural networks, and generative adversarial networks, and introduce attention mechanisms for the prediction models; Performing model training based on the constructed prediction model, including adjusting the parameters of the prediction model using an adaptive optimization algorithm; Perform feature extraction based on the trained prediction model, including performing convolution operations with the input image through the convolution kernel to extract local features of the image; The danger threshold is determined based on the output image features.

2. According to the method for detecting the construction environment of an intelligent helmet based on image recognition according to claim 1, it is characterized in that: The construction image is denoised and the denoised image data is enhanced, including using a median filter to replace the grayscale value of each pixel in the image data with the median of the grayscale values ​​in the neighborhood of the pixel, then calculating the grayscale histogram of the original image through histogram equalization, calculating the cumulative distribution function based on the grayscale histogram, and mapping the grayscale values ​​of the original image through the cumulative distribution function to obtain an enhanced image.

3. The method for detecting construction environment of an intelligent safety helmet based on image recognition according to claim 2 is characterized in that: The method comprises constructing a prediction model based on a convolutional neural network, a recurrent neural network and a generative adversarial network, comprising constructing a convolutional neural network composed of a plurality of convolutional layers, an activation function layer, a pooling layer and a fully connected layer, wherein the convolutional layer is composed of a plurality of convolutional kernels of different sizes, and the input image data is passed through a plurality of convolutional layers to obtain a set of feature maps, and the different feature maps output by the convolutional layer are input to the activation function layer for nonlinear transformation, and then enter the pooling layer for downsampling operation, and finally the fully connected layer performs feature integration.

4. The method for detecting construction environment of an intelligent safety helmet based on image recognition according to claim 3 is characterized in that: The prediction model based on convolutional neural network, recurrent neural network and generative adversarial network is constructed, and the recurrent neural network adopts long short-term memory network, the input gate of the recurrent neural network is used to receive feature information output by the fully connected layer, and the output gate of the recurrent neural network transmits the output image data information to the generator of the generative adversarial network. The realistic construction environment image sequence generated by the adversarial network is used as additional training data and input into the recurrent neural network for training.

5. The method for detecting construction environment of an intelligent helmet based on image recognition according to claim 4 is characterized in that: The attention mechanism is introduced into the prediction model, including obtaining two feature maps through two convolutional layers for the input feature map respectively, and then performing element-by-element multiplication and sigmoid activation on the two feature maps to obtain a spatial attention map, and finally, performing element-by-element multiplication of the input feature map and the spatial attention map to obtain an enhanced feature map.

6. The method for detecting construction environment of an intelligent helmet based on image recognition according to claim 5 is characterized in that: The method of adjusting the parameters of the model using the adaptive optimization algorithm includes calculating the exponential moving average of the square of the model gradient according to each parameter in the prediction model, and updating the model parameters using the exponential moving average. The calculation formula of the exponential moving average is: ， in, It is expressed as the exponential moving average of the square of the gradient at time t, It is expressed as a decay rate, which is a value between 0 and 1. It is expressed as the exponential moving average of the square of the gradient at the previous moment, It is expressed as the gradient at the current moment, that is, the derivative of the loss function with respect to the current parameters.

7. The method for detecting construction environment of an intelligent helmet based on image recognition according to claim 6 is characterized in that: The method of judging the danger threshold based on the output image features includes dividing the image feature results output by the prediction model into dangerous equipment features, personnel behavior features and environmental condition features, assigning weights to different features to build a danger assessment model, and using the output structure of the danger assessment model to judge the danger threshold.

8. An intelligent safety helmet construction environment detection system based on image recognition, characterized in that: include: The data acquisition module is configured to: acquire image data of the construction environment; The preprocessing module is configured to: preprocess the acquired image data, including denoising the construction image, and performing image enhancement on the denoised image data; The conversion module is configured to: use the preprocessed image as input to build a prediction model, including building a prediction model based on a convolutional neural network, a recurrent neural network, and a generative adversarial network, and introduce an attention mechanism for the prediction model; The model module is configured to: perform model training according to the constructed prediction model, including adjusting the parameters of the prediction model by using an adaptive optimization algorithm; The feature extraction module is configured to: perform feature extraction based on the trained prediction model, including performing a convolution operation with the input image through a convolution kernel to extract local features of the image; The transformation module is configured to: determine the danger threshold according to the output image features.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded and executed by a processor of a terminal device according to a method for detecting a construction environment of an intelligent safety helmet based on image recognition as claimed in claim 1.

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method for detecting construction environment of an intelligent safety helmet based on image recognition as claimed in claim 1.

Images