Power transmission line risk identification system and method based on large cloud model

Abstract

The invention discloses a power transmission line risk identification system and method based on a cloud large model, and the system comprises a cloud platform which is used for pulling a video stream from a corresponding monitoring device according to the identification result of an image identification module, and transmitting the identification result of a video identification module to a large model identification module; the image identification module is used for carrying out risk identification on the alarm image returned by the monitoring and photographing equipment; the video identification module is used for carrying out risk behavior identification on the video stream forwarded by the cloud platform through a pre-trained target trajectory tracking model in the video identification module to obtain an identification result; and the cloud identification module is used for summarizing risk behaviors existing in the identification result of the video identification module forwarded by the cloud platform into a natural language through a risk identification large model pre-trained in the cloud identification module so as to obtain specific risk behaviors. According to the method, the power transmission line risk identification capability can be improved, and omission caused by personnel negligence can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of power transmission line inspection technology, and in particular to a power transmission line risk identification system and method based on a cloud-based large model. Background Art

[0002] At present, the inspection of power transmission lines mainly relies on monitoring equipment instead of regular inspections by personnel. In the monitoring room, there needs to be a dedicated person on duty to monitor the pictures and videos sent back by the monitoring equipment, and process the alarm information generated by the monitoring equipment through its own AI recognition of special vehicles and other situations. In order to ensure the safety of transmission lines, it is necessary to obtain the on-site conditions of transmission lines in real time. However, due to the long distance and wide distribution of transmission lines, it is impossible to record 7*24h like the monitoring system, which will cause a lot of hard disk, traffic, and power consumption. In order to solve this problem, the conventional solution is to make the monitoring equipment run in low power consumption mode, and regularly capture pictures and send them back to the power platform. The timing is usually set from 10 minutes to 60 minutes. This capture method can effectively ensure the grasp of the on-site conditions of the transmission lines, but there will be a vacuum period. At the same time, a large number of captured pictures still need to be screened by manpower, and it is difficult to find risks in time.

[0003] Since the computing power of monitoring equipment is very limited, the existing technology can only perform simple risk identification through its built-in AI chip. Limited by the chip computing power, in order to meet the real-time requirements, only a smaller model network can be used. The detection effect of the small model is poor, and there are many false alarms and missed alarms. After detecting a potential risk target, the monitoring equipment cannot determine whether it constitutes a risk by itself, and still needs to trigger an alarm and upload it to the platform for manual secondary processing. At the same time, due to the principle of "prefer false alarms to missed alarms" in power transmission line inspections, the risk identification threshold is usually set low in existing solutions, resulting in a large number of false alarms and invalid alarms in monitoring equipment. In addition, some targets will last for a period of time from entering the screen to leaving the screen. During this period of time, the target will be continuously detected and multiple alarms will be generated, resulting in a waste of traffic and hard disk space. In addition, manual secondary processing is required, which wastes human resources for secondary inspection. However, the actual risks are often hidden in a large amount of data, and manual inspections are prone to omissions, which can easily cause significant losses.

[0004] Therefore, in order to process the data sent back by the monitoring equipment, eliminate interference and find out potential risks, it is necessary to design a transmission line risk identification system and method to replace manual secondary analysis and further reduce manpower work. Summary of the invention

[0005] Purpose of the invention: In view of the above-mentioned shortcomings, the present invention proposes a transmission line risk identification system and method based on a cloud-based large model, which can not only improve the detection rate and accuracy of the initial inspection on the equipment side and improve the transmission line risk identification capability, but also reduce omissions caused by human negligence, thereby improving the overall efficiency of transmission line risk monitoring without increasing additional costs for users.

[0006] Technical solution:

[0007] The present invention provides a transmission line risk identification system based on a cloud-based large model, comprising:

[0008] The cloud platform is used to pull the video stream from the corresponding monitoring device according to the recognition result of the image recognition module, and forward it to the video recognition module, and at the same time forward the recognition result of the video recognition module to the large model recognition module;

[0009] The image recognition module is used to receive the warning images sent back by the monitoring equipment, identify the risks through the pre-trained risk target detection model, and obtain the images containing risks;

[0010] The video recognition module is used to identify risky behaviors in the video stream forwarded by the cloud platform through the pre-trained target trajectory tracking model to obtain the recognition result;

[0011] The cloud recognition module is used to summarize the risk behaviors existing in the recognition results of the video recognition module forwarded by the cloud platform into natural language through the pre-trained risk recognition model within it, so as to obtain specific risk behaviors.

[0012] Specifically, the risk target detection model pre-trained in the image recognition module adopts the YOLO target detection model.

[0013] More specifically, during the training process of the YOLO target detection model, an adaptive directional pooling layer operator is introduced for optimization, as follows:

[0014] During the training process of the YOLO target detection model, the aspect ratio of the target is calculated based on its width and height, and the aspect ratio is compared with a set threshold;

[0015] If the aspect ratio exceeds the set threshold, the target is a slender target. If the target direction is horizontal, the pooling operator used is a 4*2 matrix, and the convolution calculation result is pooled. On the contrary, if the target direction is vertical, the pooling operator used is a 2*4 matrix, and the convolution calculation result is pooled.

[0016] The loss function of the YOLO target detection model adopts

[0017] Specifically, the target trajectory tracking model is trained by acquiring video streams of common risk behaviors. The input is the video stream, and the output is the matching degree between the motion trajectory of each risk object and the risk behavior. A threshold is set. If the matching degree exceeds the set threshold, the motion trajectory of the risk object is considered to be a risk behavior.

[0018] More specifically, the target trajectory tracking model is provided with a BoT-SORT trajectory tracking algorithm to judge and predict the direction and speed of the target's movement, and to judge whether the target's movement trajectory is a risky movement trajectory accordingly.

[0019] More specifically, a summary processing module is also provided in the video recognition module, and the summary processing module predicts possible repeated alarms within a set time in the future based on the risk motion trajectory and the corresponding speed obtained by the target trajectory tracking model, and suppresses them.

[0020] Specifically, the risk identification big model in the cloud identification module adopts the LLaVA multimodal big model that is fine-tuned and trained using the QLoRA technology.

[0021] Specifically, after the image recognition module identifies the alarm images to exclude false alarms, the cloud platform pulls the video stream from the corresponding monitoring device according to the remaining images and forwards it to the video recognition module.

[0022] Specifically, the cloud platform is built by supporting the electric power I1 protocol and receives the alarm image that complies with the I1 protocol and is sent by the monitoring device.

[0023] The present invention also provides a transmission line risk identification method using the above-mentioned transmission line risk identification system, comprising the steps of:

[0024] S1. The monitoring device monitors the images of the power transmission line, identifies the alarm images therein, and transmits the alarm images back to the image recognition module.

[0025] S2. The image recognition module uses its risk target detection model to identify the risk of the warning images sent back by the monitoring equipment, thereby screening out the images with risks;

[0026] S3: The cloud platform obtains the risky image obtained in S2, pulls the video stream from the corresponding monitoring device, and forwards it to the video recognition module;

[0027] S4, the video recognition module uses the target trajectory tracking model in it to identify risky behaviors in the video stream forwarded by the cloud platform, and uses the video with risky behaviors as its recognition result;

[0028] S5. The cloud platform forwards the recognition result of the video recognition module in S4 to the cloud recognition module, and summarizes the risk behaviors in the recognition result of the video recognition module forwarded by the cloud platform into natural language through the risk recognition model therein to obtain specific risk behaviors.

[0029] Beneficial effects: The present invention is easy to deploy, easy to use, and can be automatically executed, thereby effectively reducing the manpower burden, reducing losses caused by human negligence, reducing labor costs, and having a high risk identification rate. The present invention can also improve the accuracy of end-side target detection, effectively reduce missed alarms and false alarms, screen effective alarms, and improve the efficiency of transmission line risk monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is an example diagram of the architecture of the power transmission line risk identification system based on the cloud-based large model of the present invention;

[0031] Figure 2 It is a schematic diagram of the risk target detection model of the present invention. DETAILED DESCRIPTION

[0032] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0033] The architecture example of the transmission line risk identification system based on the cloud large model of the present invention is as follows: Figure 1 As shown, including:

[0034] The cloud platform is used to pull the video stream to the corresponding monitoring device according to the recognition result of the image recognition module, and forward it to the video recognition module, and at the same time forward the recognition result of the video recognition module to the large model recognition module.

[0035] In the present invention, the cloud platform is constructed by supporting the electric power I1 protocol and receives the alarm images that comply with the I1 protocol and are sent by the monitoring equipment.

[0036] In the present invention, after the image recognition module of the cloud platform has identified and excluded a large number of alarm images falsely reported by monitoring and shooting devices, it pulls the video stream from the corresponding monitoring and shooting devices according to the remaining images and forwards it to the video recognition module.

[0037] In the present invention, the monitoring device provides AI computing power, video review function and network communication capability on the device side. A trained model is deployed therein to identify risk targets for transmission lines. It can communicate with the cloud platform through the Internet of Things, respond to the cloud platform's request to pull video streams, and then transmit the corresponding video streams to the cloud platform.

[0038] In the present invention, a power platform can also be set up. The power platform is connected to the monitoring device through the power protocol. The monitoring device runs in a low power consumption mode and regularly captures pictures and transmits them back to the power platform. The power platform is also connected to the cloud platform and can receive alarm information sent by the cloud platform, so that the regular capture mode can be combined with the system of the present invention, integrating different monitoring methods and corresponding alarm information.

[0039] The image recognition module is provided with a pre-trained risk target detection model. In the present invention, the risk target detection model adopts the YOLO target detection model fine-tuned for the power transmission line scene, which is used to receive the warning image sent back by the monitoring device and perform risk identification, thereby identifying the image in which there is risk, so as to complete the screening of the warning image sent back by the monitoring device. Compared with the low-precision target recognition performed by the recognition model built into the monitoring device, the parameter quantity of the risk detection model used in the image recognition module can be greatly improved, and then the resolution of the input image can be correspondingly increased, and there is no need to compress the resolution of the input image like the recognition model built into the monitoring device, so that the accuracy of image recognition and the detection accuracy of small objects are greatly improved, and then the warning image recognized by the monitoring device can be screened to further identify risk targets.

[0040] In the present invention, the fine-tuning of the YOLO target detection model is specifically as follows:

[0041] Based on the training of the YOLO target detection model, an adaptive directional pooling layer operator is introduced for optimization, as shown in 2, where C3, P4, and P5 are the results of operations at different scales. As shown in the figure, the output result C3 is obtained after the input image passes through multiple convolution layers, the output result P4 is obtained after C3 passes through the convolution layer, directional pooling layer and upsampling, and the output result P5 is obtained after P4 passes through the convolution layer, directional pooling layer and upsampling. The details are as follows:

[0042] During the iterative training process, the aspect ratio of the target is calculated based on its width and height, and its aspect ratio is compared with the set threshold; if the aspect ratio exceeds the set threshold, it means that the target is a slender target. In this case, different pooling operators are used for pooling in combination with the predicted target direction, that is, the length direction of the target. Specifically, if the target direction is horizontal, the pooling operator used is a 4*2 matrix, and the convolution calculation results are pooled; conversely, if the target direction is vertical, the pooling operator used is a 2*4 matrix, and the convolution calculation results are pooled. The existing YOLO target detection model is improved by using this directional pooling operator that combines the characteristics of the target's aspect ratio. Experimental results show that the recognition accuracy of slender targets can be effectively improved, and the recognition accuracy of slender targets in some scenes can be improved by more than 15%.

[0043] In addition, in the present invention, the loss function of the YOLO target detection model adopts Specifically:

[0044]

[0045] Where IOU represents the intersection-over-union ratio between the predicted box of the YOLO target detection model and its corresponding pre-obtained true box, d represents the distance between the center point of the predicted box and the center point of the true box, c represents the diagonal distance of the minimum circumscribed rectangle of the image formed by the predicted box and the true box, V represents the correction factor, and α represents the weight function;

[0046]

[0047] Where w and h represent the width and height of the real box, w gt and h gt Represents the width and height of the prediction box.

[0048] The present invention adopts the YOLO target detection model fine-tuned for the transmission line scenario, can optimize the training of slender targets, use a targeted model training method, introduce the loss function CIOU suitable for detecting long strip targets, and introduce a directional maximum pooling layer, so as to improve the training and detection effects of slender targets such as tower cranes and cranes in the transmission line scenario. At the same time, the reasoning process of the YOLO target detection model is optimized, and the added directional pooling layer is used to extract features of the image to be detected. When the prediction box for the slender target is used during prediction, the detection rate of the slender target can be improved, which better meets the transmission line scenario.

[0049] The video recognition module is provided with a pre-trained target trajectory tracking model, which is trained by acquiring video streams of common risk behaviors of engineering vehicles, pedestrians, etc., and can then identify the risky motion trajectories of risky objects such as engineering vehicles, pedestrians, and power transmission lines in the video stream. The input is the video stream, and the output is the matching degree between the motion trajectory of each risky object and the risky behavior, and a threshold is set. If the matching degree exceeds the set threshold, the motion trajectory of the risky object is considered to be a risky behavior. In the present invention, common risky behaviors may include fast movement of risky objects such as engineering vehicles or pedestrians, large shaking of risky objects such as power transmission lines, and approach of risky objects such as engineering vehicles or pedestrians to power transmission lines.

[0050] In the present invention, the BoT-SORT trajectory tracking algorithm is set in the target trajectory tracking model to judge and predict the direction and speed of the target movement, and judge whether the target movement trajectory is a risky movement trajectory; for vehicles whose movement trajectory is far away from the transmission line, the alarm can be suppressed according to user needs, and the vehicles running at high speed or heading towards the transmission line and the special engineering vehicles at work are regarded as potential risks and generate alarms. If the target to be detected is a person, behavior recognition based on Temporal 3D ConvNets can be further used to detect high-risk behaviors such as falling, fishing, climbing, etc., so as to detect people around the transmission line.

[0051] In the present invention, the matching degree between the motion trajectory of each risk object and the risk behavior can be designed as a matching score. The target trajectory tracking model is preset with matching scores corresponding to the different motion trajectories of each risk object. If the risk object moves toward the center of the picture, the matching score is increased. If it moves away from the center of the picture, the matching score is decreased. Based on this, the corresponding matching score is calculated according to the motion trajectory of each risk object identified, and then the matching degree between the motion trajectory of each risk object and the risk behavior is obtained.

[0052] Furthermore, in addition to presetting the matching scores for different motion trajectories of different risky objects, the different moving speeds corresponding to the motion trajectories of the risky objects are also presetting corresponding matching scores.

[0053] For example, a pedestrian moving slowly towards the picture will be given 20 points, a vehicle moving at high speed towards the picture will be given 50 points, and a construction vehicle working near the center of the picture will be given 80 points. In different actual scenarios, high thresholds can be set to filter out unimportant risks, or low thresholds can be set to ensure personnel safety, etc.

[0054] Furthermore, in the present invention, a summary processing module is also provided in the video recognition module, which predicts possible repeated alarms within a set time in the future based on the risk motion trajectory and the corresponding speed obtained by the target trajectory tracking model, and suppresses them to prevent a large number of repeated alarms.

[0055] The present invention uses the target trajectory tracking model in the video recognition module to identify risky behaviors in the video stream forwarded by the cloud platform. By tracking the motion trajectory of risky objects in the video stream, such as engineering vehicles and pedestrians, it is identified whether the corresponding risky objects pose a risk to the transmission line, and the video containing risky behaviors is used as the identification result.

[0056] The cloud recognition module has a pre-trained risk recognition model, which is used to further analyze the recognition results of the video recognition module forwarded by the cloud platform, summarize the risk behaviors in the video into natural language, and identify potential risks on the transmission line;

[0057] In the present invention, the risk identification large model in the cloud identification module can adopt the LLaVA multimodal large model, and fine-tune the training by collecting existing transmission line risk cases as training samples, so that it performs better in the transmission line risk scenario. The present invention uses QLoRA technology to perform fine-tuning training to improve the hardware resource consumption, training speed, and reasoning speed during the fine-tuning process, as follows:

[0058] Normal Float (NF) is used to compress the model parameters of the LLaVA multimodal large model. Quantization using NF requires a smaller block size, but this will cause a lot of memory overhead. Therefore, double quantization (DQ) is introduced, that is, the quantization constants obtained after the aforementioned compression are secondary quantized to save memory, which will not cause significant performance degradation. At the same time, during the fine-tuning training process, ChatGPT4 is used to automatically process the data input into the LLaVA multimodal large model, and the generated results are saved as the standard answer as a verification data set. Based on the verification data set, ChatGPT is used periodically to evaluate the fine-tuning results to confirm the training results. The introduction of ChatGPT4 greatly improves the efficiency and effectiveness of fine-tuning training and reduces the labor cost when facing large amounts of data.

[0059] The present invention adopts the LLaVA multimodal large model to summarize the risk behaviors existing in the recognition results of the video recognition module forwarded by the cloud platform into natural language to obtain specific risk behaviors.

[0060] The intelligent early warning module generates alarm information based on the recognition results of the large model recognition module and uploads the alarm information to the user end.

[0061] The transmission line risk identification system based on the cloud-based large model of the present invention can retain the timed snapshot while the monitoring equipment can choose to turn on target identification, analyze the on-site situation, and actively report and indicate the type of hidden danger when engineering vehicles, smoke, wildfires, etc. are found. This greatly improves the efficiency of transmission line monitoring and the early warning function of potential risks. By using cloud-based automated filtering and detection, the present invention can reduce the dependence of risk identification tasks on personnel, improve overall efficiency, and avoid transmission line losses caused by human factors. At the same time, by using a trajectory tracking detection method, by analyzing the target intention, a more accurate risk level segmentation is achieved, further improving the efficiency of risk warning tasks. Furthermore, the present invention has a faster training convergence speed and reasoning detection rate for slender targets in transmission lines without increasing hardware costs through the improved YOLO target detection model.

[0062] The present invention also provides a transmission line risk identification method based on a cloud-based large model, comprising the steps of:

[0063] S1. The monitoring device monitors the image of the transmission line, obtains the alarm image through its built-in recognition model, and transmits the alarm image back to the image recognition module.

[0064] S2, the image recognition module uses the pre-trained risk target detection model in it to identify the risk of the warning images sent back by the monitoring equipment, so as to screen out the images with risks;

[0065] In the present invention, the risk target detection model pre-trained in the image recognition module adopts the YOLO target detection model fine-tuned for the power transmission line scenario.

[0066] S3: The cloud platform obtains the risky image obtained in S2, pulls the video stream from the corresponding monitoring device, and forwards it to the video recognition module;

[0067] S4, the video recognition module uses the pre-trained target trajectory tracking model in the video stream forwarded by the cloud platform to identify risky behaviors, and uses the video with risky behaviors as its recognition result;

[0068] In the present invention, the target trajectory tracking model is trained with the common risk behaviors of engineering vehicles, pedestrians, etc. obtained in the prior art, and then the risky motion trajectories of risky objects such as engineering vehicles, pedestrians, and power transmission lines in the video stream can be identified. The input is the video stream, and the output is the matching degree between the motion trajectory of each risk object and the risky behavior, and a threshold is set. If the matching degree exceeds the set threshold, the motion trajectory of the risky object is considered to constitute a risk.

[0069] S5. The cloud platform forwards the recognition results of the video recognition module in S4 to the cloud recognition module, and further analyzes the recognition results of the video recognition module forwarded by the cloud platform through the pre-trained risk recognition model in it, and summarizes the behaviors in the video into natural language to identify potential risks on the transmission line.

[0070] In the present invention, the risk identification large model pre-trained in the cloud identification module is trained by collecting existing transmission line risk cases as training samples, judging the potential risk of the video recognition results to the transmission lines, summarizing the risk behaviors in the video into natural language, quantifying the risk level, and describing specific risk behaviors.

[0071] The present invention also includes step S6: the intelligent early warning module generates warning information according to the recognition result of the large model recognition module, and uploads it to the user end.

[0072] The present invention is easy to deploy, easy to use, and can be automatically executed, thereby effectively reducing the manpower burden, reducing losses caused by human negligence, and reducing labor costs. Without increasing hardware costs, it can have a faster training convergence speed and reasoning detection rate for slender targets in transmission lines, and has a high risk identification rate. The present invention can quantify the degree of risk, and by analyzing the target intention, achieve more accurate risk level segmentation, further improving the efficiency of risk warning tasks. At the same time, the present invention uses cloud-based automated filtering and detection, which can reduce the dependence of risk identification tasks on personnel, improve overall efficiency, and avoid transmission line losses caused by human factors.

[0073] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations (such as quantity, shape, position, etc.) can be made to the technical scheme of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. Transmission line risk identification system based on cloud-based large model, It is characterized in that include: The cloud platform is used to pull the video stream from the corresponding monitoring device according to the recognition result of the image recognition module, and forward it to the video recognition module, and at the same time forward the recognition result of the video recognition module to the large model recognition module; The image recognition module is used to receive the warning images sent back by the monitoring equipment, identify the risks through the pre-trained risk target detection model, and obtain the images containing risks; The video recognition module is used to identify risky behaviors in the video stream forwarded by the cloud platform through the pre-trained target trajectory tracking model to obtain the recognition result; The cloud recognition module is used to summarize the risk behaviors existing in the recognition results of the video recognition module forwarded by the cloud platform into natural language through the pre-trained risk recognition model within it, so as to obtain specific risk behaviors.

2. The transmission line risk identification system according to claim 1, It is characterized in that The risk target detection model pre-trained in the image recognition module adopts the YOLO target detection model.

3. The transmission line risk identification system according to claim 2, It is characterized in that During the training process of the YOLO target detection model, an adaptive directional pooling layer operator is introduced for optimization, as follows: During the training process of the YOLO target detection model, the aspect ratio of the target is calculated based on its width and height, and the aspect ratio is compared with a set threshold; If the aspect ratio exceeds the set threshold, the target is a slender target. If the target direction is horizontal, the pooling operator used is a 4*2 matrix, and the convolution calculation result is pooled. On the contrary, if the target direction is vertical, the pooling operator used is a 2*4 matrix, and the convolution calculation result is pooled. The loss function of the YOLO target detection model adopts 4. The transmission line risk identification system according to claim 1, It is characterized in that The target trajectory tracking model is trained by acquiring a video stream of common risk behaviors. The input is the video stream, and the output is the matching degree between the motion trajectory of each risk object and the risk behavior. A threshold is set. If the matching degree exceeds the set threshold, the motion trajectory of the risk object is considered to be a risk behavior.

5. The transmission line risk identification system according to claim 4, It is characterized in that The target trajectory tracking model is provided with a BoT-SORT trajectory tracking algorithm to judge and predict the direction and speed of the target movement, and to judge whether the target movement trajectory is a risky movement trajectory accordingly.

6. The transmission line risk identification system according to claim 5, It is characterized in that The video recognition module is also provided with a summary processing module, which predicts possible repeated alarms within a set time in the future based on the risk motion trajectory and the corresponding speed obtained by the target trajectory tracking model, and suppresses them.

7. The transmission line risk identification system according to claim 1, It is characterized in that The risk identification model in the cloud identification module adopts the LLaVA multimodal model that is fine-tuned and trained using the QLoRA technology.

8. The transmission line risk identification system according to claim 1, It is characterized in that After the image recognition module identifies the alarm images to exclude false alarms, the cloud platform pulls the video stream from the corresponding monitoring device according to the remaining images and forwards it to the video recognition module.

9. The transmission line risk identification system according to claim 1, It is characterized in that The cloud platform is built by supporting the electric power I1 protocol and receives the alarm image that complies with the I1 protocol and is sent by the monitoring device.

10. A transmission line risk identification method using the transmission line risk identification system according to any one of claims 1 to 9, It is characterized in that Includes steps: S1. The monitoring device monitors the images of the power transmission line, identifies the alarm images therein, and transmits the alarm images back to the image recognition module. S2. The image recognition module uses its risk target detection model to identify the risk of the warning images sent back by the monitoring equipment, thereby screening out the images with risks; S3: The cloud platform obtains the risky image obtained in S2, pulls the video stream from the corresponding monitoring device, and forwards it to the video recognition module; S4, the video recognition module uses the target trajectory tracking model in it to identify risky behaviors in the video stream forwarded by the cloud platform, and uses the video with risky behaviors as its recognition result; S5. The cloud platform forwards the recognition result of the video recognition module in S4 to the cloud recognition module, and summarizes the risk behaviors in the recognition result of the video recognition module forwarded by the cloud platform into natural language through the risk recognition model therein to obtain specific risk behaviors.

Images