Picture classification identification and risk assessment method and system based on artificial intelligence
By using artificial intelligence identification system in the factory workshop for image classification and risk assessment, the problem of cumbersome and inaccurate manual identification is solved, and the accurate identification and risk prediction of hazardous elements in the factory workshop is achieved, and reliable risk investigation is provided. in accordance with.
Patent Information
- Application Number
- CN202410842643.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-27
- Publication Date
- 2024-11-08
AI Technical Summary
The number of images taken by cameras in the factory workshop is huge, which makes manual recognition cumbersome and inaccurate. There are errors in manual recognition, which reduces the accuracy of image recognition and cannot provide a reliable basis for risks within the factory workshop.
Using the image classification recognition and risk assessment method based on artificial intelligence, the screen elements are recognized by taking the original picture of the target place, divided into several picture sub-pictures, and the activity status data of the elements is determined by using the neural network to determine whether it is a dangerous source, and summarize it Predict the status information of dangerous events, carry out labeling of neural networks, and realize intelligent and efficient visual recognition.
It realizes accurate identification and risk prediction of hazardous elements in the factory workshop, provides reliable risk investigation basis, and improves the accuracy and efficiency of image recognition.
Smart Images

Figure CN118918353A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of artificial intelligence, and in particular to an image classification, recognition and risk assessment method and system based on artificial intelligence. Background Art
[0002] Closed spaces such as factory workshops have the characteristics of complex personnel flow and dense distribution of mechanical equipment. Once an incident such as misoperation by personnel or mechanical equipment failure occurs inside the factory workshop, it is easy to trigger a chain reaction inside the factory workshop and cause safety accidents, resulting in casualties and property losses. In order to ensure the normal and safe operation of the factory workshop, cameras are usually installed inside the factory workshop for visual monitoring. Specifically, the real-time images of the factory workshop captured by the camera will be transmitted to the monitoring background, and computer-assisted image recognition and manual image screening will be used to determine the dangerous accidents that may occur in the factory workshop. The operation of the factory workshop is complex, and the number of images captured by the camera is huge, which requires a lot of manpower and time to identify and screen the images, and it is impossible to identify the images in time. In addition, manual image recognition inevitably has errors and cannot globally identify the image screen, which reduces the accuracy of image recognition and cannot provide a basis for accurately checking the risks inside the factory workshop. Summary of the invention
[0003] In view of the defects existing in the prior art, the present invention provides an image classification, recognition and risk assessment method and system based on artificial intelligence, which takes a number of original images of different areas of a target place, and identifies the elements of the original images to obtain the element attribute information of the original images, and preliminarily identifies the elements inside the target place that may cause unexpected dangers; divides the original image into a number of image sub-images corresponding to all the elements, and identifies the activity status data of the elements through a neural network to determine whether the elements belong to a hazard source, and perform targeted identification of different types of elements; then summarizes the activity status data of all elements belonging to the hazard source through a neural network, predicts the occurrence status information of dangerous events in the target place, and accurately and comprehensively predicts the dangerous events caused by the elements; and based on the occurrence status information of the dangerous events, performs neural network labeling processing on the elements in the original image that belong to the hazard source, thereby realizing intelligent and efficient visual recognition of the target place, and ensuring that the output original image can provide a reliable basis for risk investigation.
[0004] The present invention provides an image classification, recognition and risk assessment method based on artificial intelligence, comprising the following steps: Step S1, photographing a target location to obtain a number of original images of different areas of the target location; identifying screen elements of the original images to obtain element attribute information of the original images; Step S2, based on the element attribute information, performing screen capture processing on the original image to obtain a number of image sub-images corresponding to all elements one by one; performing neural network recognition on the image sub-images to obtain activity status data of the elements in the target place; based on the activity status data, determining whether the element is a dangerous source inside the target place; Step S3, summarizing the activity status data of all elements belonging to the hazard source by neural network, and predicting the occurrence status information of the hazard event in the target place; Step S4, based on the dangerous event occurrence status information, the original image is subjected to neural network labeling processing of the elements belonging to the dangerous source; and based on the result of the neural network labeling processing, it is determined whether the original image meets the preset screen content label layout conditions, thereby adjusting the output status of the original image.
[0005] In one embodiment disclosed in the present application, in step S1, a target place is photographed to obtain a plurality of original pictures of different areas of the target place; and picture element recognition is performed on the original pictures to obtain element attribute information of the original pictures, including: Performing non-contact detection on the target location to obtain relative position information between different areas inside the target location and the shooting reference point; adjusting the shooting depth of field and shooting field of view of the corresponding area inside the target location based on the relative position information, thereby shooting the corresponding area to obtain an original picture of the corresponding area; After the original image is subjected to image noise reduction filtering and image pixel edge sharpening processing, the original image is input into the CNN model for image element recognition to obtain all elements existing on the original image and their visual attribute information as the element attribute information; wherein the elements include personnel, mechanical equipment, background environment and materials.
[0006] In one embodiment disclosed in the present application, in the step S2, based on the element attribute information, the original image is subjected to screen capture processing to obtain a plurality of image sub-images corresponding to all elements one by one; the image sub-images are subjected to neural network recognition to obtain activity status data of the elements in the target place; based on the activity status data, it is determined whether the element belongs to a danger source inside the target place, including: Based on the visual attribute information of all elements of the original picture, determine the associated picture area of each element in the original picture; based on the boundary of the associated picture area, perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to the element; based on the element types corresponding to all the picture sub-pictures, divide all the picture sub-pictures into a plurality of picture sub-picture sets; wherein all the picture sub-pictures under each picture sub-picture set correspond to the same type of elements; Input all the picture sub-pictures under the picture sub-picture set into the Bayes-FNN model for recognition, and obtain the activity state data of the elements in the target place; wherein the activity state data includes the behavior activity state data of the personnel, mechanical equipment, background environment and materials in the target place; Based on the activity status data, the influence range and duration of the element in the process of performing corresponding behavioral activities inside the target place are obtained; then based on the influence range and duration, it is determined whether the element is a dangerous source inside the target place.
[0007] In one embodiment disclosed in the present application, in step S3, the activity status data of all elements belonging to the danger source are aggregated by a neural network to predict the occurrence status information of the dangerous event at the target location, including: Based on all the behavior activity types under the activity status data of all the elements belonging to the hazardous source, the activity status data of all the elements are deduplicated; based on the generation time of the activity status data of all the elements belonging to the hazardous source, the activity status data that has completed the deduplication processing are sorted to generate a corresponding activity status data queue; The activity status data queue is input into the corresponding FNN model for summary processing to predict the occurrence status information of the dangerous event at the target location; wherein the occurrence status information of the dangerous event includes the type of dangerous event, the time of occurrence, and the scope of the affected area.
[0008] In one embodiment disclosed in the present application, in the step S4, based on the dangerous event occurrence status information, the original image is subjected to neural network labeling processing of the elements belonging to the dangerous source; and based on the result of the neural network labeling processing, it is judged whether the original image meets the preset picture content label layout condition, thereby adjusting the output state of the original image, including: Input the dangerous event occurrence status information and the personnel flow prediction data inside the target place into the GAN model for processing to predict the risk level information of the dangerous event; then based on the risk level information, label the location of the elements belonging to the dangerous source and the risk level information of the dangerous event generated in the original image; Performing picture information recognition on the original picture that has completed labeling processing to obtain the picture information distribution density of the original picture that has completed labeling processing; if the picture information distribution density is greater than a preset information distribution density threshold, then judging that the original picture meets the preset picture content label layout condition; otherwise, judging that the original picture does not meet the preset picture content label layout condition; when the original picture meets the preset picture content label layout condition, directly outputting the original picture that has completed labeling processing; when the original picture does not meet the preset picture content label layout condition, re-performing the labeling processing on the original picture until it meets the preset picture content label layout condition and then directly outputting it.
[0009] The present invention also provides an image classification recognition and risk assessment system based on artificial intelligence, including: An original picture acquisition module is used to photograph a target location and obtain a number of original pictures of different areas of the target location; An element recognition module is used to recognize the elements of the original picture to obtain the element attribute information of the original picture; An original picture segmentation module is used to perform screen capture processing on the original picture based on the element attribute information to obtain a number of picture sub-pictures corresponding to all elements one by one; A hazard source identification module is used to perform neural network recognition on the image sub-picture to obtain activity status data of the element in the target place; based on the activity status data, determine whether the element belongs to a hazard source inside the target place; A dangerous event prediction module is used to aggregate the activity status data of all elements belonging to the dangerous source through a neural network to predict the occurrence status information of the dangerous event in the target place; An image labeling processing module is used to perform neural network labeling processing of elements belonging to the danger source on the original image based on the dangerous event occurrence status information; The image output adjustment module is used to determine whether the original image meets the preset picture content label layout conditions based on the result of the neural network labeling processing, so as to adjust the output state of the original image.
[0010] In one embodiment disclosed in the present application, the original picture acquisition module is used to photograph a target location to obtain a number of original pictures of different areas of the target location, including: Performing non-contact detection on the target location to obtain relative position information between different areas inside the target location and the shooting reference point; adjusting the shooting depth of field and shooting field of view of the corresponding area inside the target location based on the relative position information, thereby shooting the corresponding area to obtain an original picture of the corresponding area; The element recognition module is used to perform screen element recognition on the original picture to obtain element attribute information of the original picture, including: After the original image is subjected to image noise reduction filtering and image pixel edge sharpening processing, the original image is input into the CNN model for image element recognition to obtain all elements existing on the original image and their visual attribute information as the element attribute information; wherein the elements include personnel, mechanical equipment, background environment and materials.
[0011] In one embodiment disclosed in the present application, the original picture segmentation module is used to perform screen capture processing on the original picture based on the element attribute information to obtain a number of picture sub-pictures corresponding to all elements one by one, including: Based on the visual attribute information of all elements of the original picture, determine the associated picture area of each element in the original picture; based on the boundary of the associated picture area, perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to the element; based on the element types corresponding to all the picture sub-pictures, divide all the picture sub-pictures into a plurality of picture sub-picture sets; wherein all the picture sub-pictures under each picture sub-picture set correspond to the same type of elements; The hazard source identification module is used to perform neural network recognition on the image sub-image to obtain activity status data of the element in the target place; based on the activity status data, determine whether the element belongs to a hazard source inside the target place, including: Input all the picture sub-pictures under the picture sub-picture set into the Bayes-FNN model for recognition, and obtain the activity state data of the elements in the target place; wherein the activity state data includes the behavior activity state data of the personnel, mechanical equipment, background environment and materials in the target place; Based on the activity status data, the influence range and duration of the element in the process of performing corresponding behavioral activities inside the target place are obtained; then based on the influence range and duration, it is determined whether the element is a dangerous source inside the target place.
[0012] In one embodiment disclosed in the present application, the dangerous event prediction module is used to aggregate the activity state data of all elements belonging to the dangerous source through a neural network to predict the dangerous event occurrence state information of the target place, including: Based on all the behavior activity types under the activity status data of all the elements belonging to the hazardous source, the activity status data of all the elements are deduplicated; based on the generation time of the activity status data of all the elements belonging to the hazardous source, the activity status data that has completed the deduplication processing are sorted to generate a corresponding activity status data queue; The activity status data queue is input into the corresponding FNN model for summary processing to predict the occurrence status information of the dangerous event at the target location; wherein the occurrence status information of the dangerous event includes the type of dangerous event, the time of occurrence and the scope of the affected area.
[0013] In one embodiment disclosed in the present application, the image labeling processing module is used to perform neural network labeling processing of elements belonging to the danger source on the original image based on the dangerous event occurrence status information, including: Input the dangerous event occurrence status information and the personnel flow prediction data inside the target place into the GAN model for processing to predict the risk level information of the dangerous event; then based on the risk level information, label the location of the elements belonging to the dangerous source and the risk level information of the dangerous event generated in the original image; The picture output adjustment module is used to determine whether the original picture meets the preset picture content label layout conditions based on the result of the neural network labeling processing, so as to adjust the output state of the original picture, including: Performing picture information recognition on the original picture that has completed labeling processing to obtain the picture information distribution density of the original picture that has completed labeling processing; if the picture information distribution density is greater than a preset information distribution density threshold, then judging that the original picture meets the preset picture content label layout condition; otherwise, judging that the original picture does not meet the preset picture content label layout condition; when the original picture meets the preset picture content label layout condition, directly outputting the original picture that has completed labeling processing; when the original picture does not meet the preset picture content label layout condition, re-performing the labeling processing on the original picture until it meets the preset picture content label layout condition and then directly outputting it.
[0014] Compared with the existing technology, the artificial intelligence-based image classification, recognition and risk assessment method and system take several original images of different areas of the target place, and identify the elements of the original images to obtain the element attribute information of the original images, and preliminarily identify the elements inside the target place that may cause unexpected dangers; the original image is divided into several image sub-screens corresponding to all elements, and the activity status data of the elements are identified through a neural network to determine whether the elements are hazardous sources, and to perform targeted identification of different types of elements; the activity status data of all elements belonging to the hazardous sources are then summarized through a neural network to predict the status information of hazardous events in the target place, and to accurately and comprehensively predict the hazardous events caused by the elements; based on the status information of hazardous events, the original image is subjected to neural network labeling processing of the elements belonging to the hazardous source, to achieve intelligent and efficient visual recognition of the target place, and to ensure that the output original image can provide a reliable basis for risk investigation.
[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.
[0016] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0018] Figure 1 A schematic diagram of the process of the artificial intelligence-based image classification, recognition and risk assessment method provided by the present invention; Figure 2 A schematic diagram of the framework of the artificial intelligence-based image classification, recognition and risk assessment system provided by the present invention. DETAILED DESCRIPTION
[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0020] See also Figure 1 , is a flow chart of an artificial intelligence-based image classification, recognition and risk assessment method provided in an embodiment of the present invention. The artificial intelligence-based image classification, recognition and risk assessment method includes: Step S1, photographing a target location to obtain a number of original images of different areas of the target location; identifying the elements of the original images to obtain element attribute information of the original images; Step S2, based on the attribute information of the element, the original image is intercepted to obtain a number of image sub-images corresponding to all the elements; neural network recognition is performed on the image sub-images to obtain the activity status data of the element in the target place; based on the activity status data, it is determined whether the element belongs to a dangerous source inside the target place; Step S3, summarizing the activity status data of all elements belonging to the hazard source through a neural network to predict the occurrence status information of the hazard event at the target location; Step S4, based on the dangerous event occurrence status information, the original image is subjected to neural network labeling processing of the elements belonging to the dangerous source; and based on the result of the neural network labeling processing, it is determined whether the original image meets the preset picture content label layout conditions, thereby adjusting the output state of the original image.
[0021] The image classification, recognition and risk assessment method based on artificial intelligence takes several original images of different areas of the target place, and identifies the elements of the original images to obtain the element attribute information of the original images, and preliminarily identifies the elements inside the target place that may cause unexpected dangers; the original image is divided into several image sub-images corresponding to all elements, and the activity status data of the elements are identified through a neural network to determine whether the elements are hazardous sources, and to perform targeted identification of different types of elements; the activity status data of all elements belonging to the hazardous sources are then summarized through a neural network to predict the status information of hazardous events in the target place, and to accurately and comprehensively predict the hazardous events caused by the elements; based on the status information of hazardous events, the original image is subjected to neural network labeling processing of the elements belonging to the hazardous source, to achieve intelligent and efficient visual recognition of the target place, and to ensure that the output original image can provide a reliable basis for risk investigation.
[0022] Preferably, in step S1, the target place is photographed to obtain a number of original pictures of different areas of the target place; the picture elements of the original pictures are identified to obtain the element attribute information of the picture of the original pictures, including: Conduct non-contact detection on the target location to obtain relative position information between different areas inside the target location and the shooting reference point; based on the relative position information, adjust the shooting depth of field and shooting field of view of the corresponding area inside the target location, thereby shooting the corresponding area to obtain an original picture of the corresponding area; After the original image is subjected to image noise reduction filtering and image pixel edge sharpening processing, the original image is input into the CNN model for image element recognition to obtain all elements existing on the original image and their visual attribute information as the element attribute information; wherein the elements include personnel, mechanical equipment, background environment and materials.
[0023] In the above technical scheme, the target places such as factory workshops include different elements such as staff, mechanical equipment, water, electricity and gas pipeline layout and production raw materials. These elements are distributed in different areas inside the target place, and the elements in each area may cause misoperation or failure, etc., resulting in accidental dangers such as explosion or fire. In order to perform global and accurate visual detection of the target place, it is necessary to take targeted photos of different areas inside the target place to provide a reliable image data basis for subsequent visual recognition. First, a scanning radar integrated with a camera is used to perform radar scanning detection on the target place, and the relative distance information and relative azimuth information between different areas inside the target place and the shooting reference point (i.e., the location point of the camera) are obtained. Based on the relative distance information and relative azimuth information, the focal length and field of view angle of the camera for shooting each area inside the target place are determined, so as to adjust the shooting focal length (i.e., the shooting depth of field) and shooting field of view angle (i.e., the shooting field of view range) of the corresponding area, so as to ensure that the original picture taken can fully record all elements of the corresponding area, and provide reliable data for subsequent recognition of the original picture. The original image is also subjected to image noise reduction filtering and image pixel edge sharpening processing, which can effectively reduce the interference noise of the original image, and then the original image is input into the CNN model for image element recognition. The convolutional neural network CNN model is used to detect and identify specific elements of the original image to obtain the personnel, mechanical equipment, background environment (such as water, electricity and gas pipeline layout), materials (such as production raw materials) and their visual attribute information on the original image, wherein the visual attribute information may be but is not limited to the pixel texture and pixel contour feature information of elements such as personnel, mechanical equipment, background environment and materials in the original image, providing a reliable basis for the subsequent segmentation of the original image.
[0024] Preferably, in step S2, based on the element attribute information, the original image is subjected to screen capture processing to obtain a plurality of image sub-images corresponding to all elements one by one; neural network recognition is performed on the image sub-images to obtain activity status data of the element in the target place; based on the activity status data, it is determined whether the element belongs to a danger source inside the target place, including: Based on the visual attribute information of all elements of the original picture, determine the associated picture area of each element in the original picture; based on the boundary of the associated picture area, perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to the element; based on the element types corresponding to all the picture sub-pictures, divide all the picture sub-pictures into several picture sub-picture sets; wherein all the picture sub-pictures under each picture sub-picture set correspond to the same type of elements; Input all the picture sub-pictures under the picture sub-picture set into the Bayes-FNN model for recognition, and obtain the activity state data of the element in the target place; wherein the activity state data includes the behavior activity state data of the personnel, mechanical equipment, background environment and materials in the target place; Based on the activity status data, the influence range and duration of the element in the process of performing corresponding behavioral activities inside the target place are obtained; then based on the influence range and duration, it is determined whether the element is a hazard source inside the target place.
[0025] In the above technical solution, the pixel textures and pixel contours of the elements such as personnel, machinery and equipment, background environment and materials in each area of the target place are not the same on the corresponding original picture. In order to perform individual and refined detection and identification of each element on the original picture, the visual attribute information such as pixel texture and pixel contour of all elements in the original picture is used as a reference to determine the associated picture area of each element in the original picture, and then the original picture is intercepted based on the boundary of the associated picture area to obtain a picture sub-picture corresponding to each element, so that each picture sub-picture contains one element, avoiding the detection and identification series interference caused by the picture sub-picture containing multiple elements at the same time. Then, based on the element types corresponding to all the picture sub-pictures, all the picture sub-pictures are divided into several picture sub-picture sets, that is, each picture sub-picture set only contains the same type of elements about personnel, machinery and equipment, background environment or materials. Then, all the picture sub-pictures under the picture sub-picture set are input into the Bayes-FNN model for identification, and the activity state data of the element in the target place is obtained, so that the behavior activities of each element in the target place during the entire operation process can be comprehensively detected and identified; wherein the activity state data may be but not limited to the actions of personnel operating mechanical equipment, the working state of mechanical equipment, the working state of water, electricity and gas pipelines in the background environment, the number and location of material accumulation, etc., and the activity state data of different elements have a direct causal relationship with whether a fire or explosion occurs in the area within the target place. Based on the activity state data, the influence range size and duration length of the element in the corresponding behavior activity process within the target place are obtained, such as the interference influence range size and duration length of the element's behavior activity in the corresponding area during the behavior activity process. If the influence range is greater than the preset range area threshold or the duration length is greater than the preset time length threshold, it is determined that the element belongs to the hazard source inside the target place; otherwise, it is determined that the element does not belong to the hazard source inside the target place, and the assessment object is accurately defined for the subsequent risk assessment of the target place.
[0026] Preferably, in step S3, the activity status data of all elements belonging to the hazard source are summarized by a neural network to predict the occurrence status information of the hazard event at the target location, including: Based on all the behavior activity types under the activity status data of all the elements belonging to the hazardous source, the activity status data of all the elements are deduplicated; based on the generation time of the activity status data of all the elements belonging to the hazardous source, the activity status data that has completed the deduplication processing are sorted to generate a corresponding activity status data queue; The activity status data queue is input into the corresponding FNN model for aggregation and processing to predict the occurrence status information of the dangerous event at the target location; wherein the occurrence status information of the dangerous event includes the type of dangerous event, the time of occurrence and the scope of the affected area.
[0027] In the above technical solution, the activity status data of all elements belonging to the hazardous source may overlap and be repeated. In order to reduce the processing workload of the activity status data, the activity status data of all elements are deduplicated based on all behavioral activity types under the activity status data of all elements belonging to the hazardous source, thereby reducing the noise interference of the activity status data. Then, based on the generation time of the activity status data of all elements belonging to the hazardous source from early to late or from late to early, the activity status data that has completed the data deduplication processing is sorted and a corresponding activity status data queue is generated, which can improve the orderliness of the activity status data. The activity status data queue is also input into the corresponding FNN model for summary processing to predict the type, time and scope of the hazardous event at the target location, thereby improving the credibility of the hazardous event prediction.
[0028] Preferably, in step S4, based on the dangerous event occurrence status information, the original image is subjected to neural network labeling processing of the elements belonging to the dangerous source; and based on the result of the neural network labeling processing, it is judged whether the original image meets the preset picture content label layout condition, so as to adjust the output state of the original image, including: The dangerous event occurrence status information and the personnel flow prediction data within the target location are input into the GAN model for processing to predict the risk level information of the dangerous event; based on the risk level information, the original image is labeled with the location of the elements belonging to the dangerous source and the risk level information of the dangerous event generated; Performing picture information recognition on the original picture that has completed labeling processing to obtain the picture information distribution density of the original picture that has completed labeling processing; if the picture information distribution density is greater than a preset information distribution density threshold, then judging that the original picture meets the preset picture content label layout conditions; otherwise, judging that the original picture does not meet the preset picture content label layout conditions; when the original picture meets the preset picture content label layout conditions, directly outputting the original picture that has completed labeling processing; when the original picture does not meet the preset picture content label layout conditions, re-performing the labeling processing on the original picture until it meets the preset picture content label layout conditions and then directly outputting it.
[0029] In the above technical scheme, if an element inside the target place is exposed to an unexpected dangerous event such as fire or explosion, it will pose a threat to the personnel in the corresponding range inside the target place, thereby generating dangerous events of different high and low risk levels. To this end, the dangerous event occurrence status information and the personnel flow prediction data inside the target place are input into the GAN model for processing to predict the risk level information of the dangerous event; then based on the risk level information, the original image is labeled with the position of the element belonging to the dangerous source and the risk level information of the dangerous event generated, so that the dangerous events that will occur inside the target place can be visualized. In addition, the original image that has been labeled is subjected to picture information recognition to obtain the picture information distribution density of the original image that has been labeled, and the picture information distribution density is compared with a threshold value to determine whether the original image meets the preset picture content label layout conditions, thereby adjusting the output state of the original image, so that the output original image can provide a reliable basis for risk investigation.
[0030] See also Figure 2 , is a schematic diagram of the framework of an artificial intelligence-based image classification, recognition and risk assessment system provided in an embodiment of the present invention. The artificial intelligence-based image classification, recognition and risk assessment system includes: The original picture acquisition module is used to shoot the target place and obtain a number of original pictures of different areas of the target place; An element recognition module is used to recognize the elements of the original image and obtain the element attribute information of the original image; The original picture segmentation module is used to perform screen capture processing on the original picture based on the element attribute information to obtain a number of picture sub-pictures corresponding to all the elements one by one; The hazard source identification module is used to perform neural network recognition on the image sub-image to obtain the activity state data of the element in the target place; based on the activity state data, it is determined whether the element belongs to the hazard source inside the target place; The dangerous event prediction module is used to aggregate the activity status data of all elements belonging to the dangerous source through a neural network to predict the occurrence status information of the dangerous event in the target place; An image labeling processing module is used to perform neural network labeling processing of elements belonging to the danger source on the original image based on the occurrence status information of the danger event; The image output adjustment module is used to determine whether the original image meets the preset picture content label layout conditions based on the result of the neural network labeling processing, so as to adjust the output state of the original image.
[0031] The AI-based image classification, recognition and risk assessment system takes several original images of different areas of the target site, identifies the elements in the original images, obtains the element attribute information of the original images, and preliminarily identifies the elements inside the target site that may cause unexpected dangers; divides the original image into several image sub-screens corresponding to all elements, and uses a neural network to identify the activity status data of the elements, determines whether the elements are hazardous sources, and performs targeted identification of different types of elements; then summarizes the activity status data of all elements belonging to hazardous sources through a neural network, predicts the status information of hazardous events in the target site, and accurately and comprehensively predicts hazardous events caused by the elements; and also performs neural network labeling processing on the elements in the original image that are hazardous sources based on the status information of hazardous events, to achieve intelligent and efficient visual recognition of the target site, and ensure that the output original image can provide a reliable basis for risk investigation.
[0032] Preferably, the original picture acquisition module is used to photograph the target place to obtain a number of original pictures of different areas of the target place, including: Conduct non-contact detection on the target location to obtain relative position information between different areas inside the target location and the shooting reference point; based on the relative position information, adjust the shooting depth of field and shooting field of view of the corresponding area inside the target location, thereby shooting the corresponding area to obtain an original picture of the corresponding area; The element recognition module is used to recognize the elements of the original picture to obtain the element attribute information of the original picture, including: After the original image is subjected to image noise reduction filtering and image pixel edge sharpening processing, the original image is input into the CNN model for image element recognition to obtain all elements existing on the original image and their visual attribute information as the element attribute information; wherein the elements include personnel, mechanical equipment, background environment and materials.
[0033] In the above technical scheme, the target places such as factory workshops include different elements such as staff, mechanical equipment, water, electricity and gas pipeline layout and production raw materials. These elements are distributed in different areas inside the target place, and the elements in each area may cause misoperation or failure, etc., resulting in accidental dangers such as explosion or fire. In order to perform global and accurate visual detection of the target place, it is necessary to take targeted photos of different areas inside the target place to provide a reliable image data basis for subsequent visual recognition. First, a scanning radar integrated with a camera is used to perform radar scanning detection on the target place, and the relative distance information and relative azimuth information between different areas inside the target place and the shooting reference point (i.e., the location point of the camera) are obtained. Based on the relative distance information and relative azimuth information, the focal length and field of view angle of the camera for shooting each area inside the target place are determined, so as to adjust the shooting focal length (i.e., the shooting depth of field) and shooting field of view angle (i.e., the shooting field of view range) of the corresponding area, so as to ensure that the original picture taken can fully record all elements of the corresponding area, and provide reliable data for subsequent recognition of the original picture. The original image is also subjected to image noise reduction filtering and image pixel edge sharpening processing, which can effectively reduce the interference noise of the original image, and then the original image is input into the CNN model for image element recognition. The convolutional neural network CNN model is used to detect and identify specific elements of the original image to obtain the personnel, mechanical equipment, background environment (such as water, electricity and gas pipeline layout), materials (such as production raw materials) and their visual attribute information on the original image, wherein the visual attribute information may be but is not limited to the pixel texture and pixel contour feature information of elements such as personnel, mechanical equipment, background environment and materials in the original image, providing a reliable basis for the subsequent segmentation of the original image.
[0034] Preferably, the original picture segmentation module is used to perform screen capture processing on the original picture based on the element attribute information to obtain a number of picture sub-pictures corresponding to all elements one by one, including: Based on the visual attribute information of all elements of the original picture, determine the associated picture area of each element in the original picture; based on the boundary of the associated picture area, perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to the element; based on the element types corresponding to all the picture sub-pictures, divide all the picture sub-pictures into several picture sub-picture sets; wherein all the picture sub-pictures under each picture sub-picture set correspond to the same type of elements; The hazard source identification module is used to perform neural network recognition on the image sub-image to obtain activity status data of the element in the target place; based on the activity status data, determine whether the element is a hazard source inside the target place, including: Input all the picture sub-pictures under the picture sub-picture set into the Bayes-FNN model for recognition, and obtain the activity state data of the element in the target place; wherein the activity state data includes the behavior activity state data of the personnel, mechanical equipment, background environment and materials in the target place; Based on the activity status data, the influence range and duration of the element in the process of performing corresponding behavioral activities inside the target place are obtained; then based on the influence range and duration, it is determined whether the element is a hazard source inside the target place.
[0035] In the above technical solution, the pixel textures and pixel contours of the personnel, mechanical equipment, background environment, materials and other elements in each area of the target place are not the same on the corresponding original picture. In order to perform individual and refined detection and identification of each element on the original picture, the pixel texture and pixel contour and other visual attribute information of all elements in the original picture are used as a reference to determine the associated picture area of each element in the original picture, and then the boundary of the associated picture area is used as a reference to perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to each element, so that each picture sub-picture contains one element, avoiding the detection and identification series interference caused by the picture sub-picture containing multiple elements at the same time. Then, based on the element types corresponding to all the picture sub-pictures, all the picture sub-pictures are divided into several picture sub-picture sets, that is, each picture sub-picture set only contains the same type of elements about personnel, mechanical equipment, background environment or materials. Then, all the picture sub-pictures under the picture sub-picture set are input into the Bayes-FNN model for identification, and the activity state data of the element in the target place is obtained, so that the behavior activities of each element in the target place during the entire operation process can be comprehensively detected and identified; wherein the activity state data may be but not limited to the actions of personnel operating mechanical equipment, the working state of mechanical equipment, the working state of water, electricity and gas pipelines in the background environment, the number and location of material accumulation, etc., and the activity state data of different elements have a direct causal relationship with whether a fire or explosion occurs in the area within the target place. Based on the activity state data, the influence range size and duration length of the element in the corresponding behavior activity process within the target place are obtained, such as the interference influence range size and duration length of the element's behavior activity in the corresponding area during the behavior activity process. If the influence range is greater than the preset range area threshold or the duration length is greater than the preset time length threshold, it is determined that the element belongs to the hazard source inside the target place; otherwise, it is determined that the element does not belong to the hazard source inside the target place, and the assessment object is accurately defined for the subsequent risk assessment of the target place.
[0036] Preferably, the dangerous event prediction module is used to aggregate the activity status data of all elements belonging to the dangerous source through a neural network to predict the dangerous event occurrence status information of the target place, including: Based on all the behavior activity types under the activity status data of all the elements belonging to the hazardous source, the activity status data of all the elements are deduplicated; based on the generation time of the activity status data of all the elements belonging to the hazardous source, the activity status data that has completed the deduplication processing are sorted to generate a corresponding activity status data queue; The activity status data queue is input into the corresponding FNN model for aggregation and processing to predict the occurrence status information of the dangerous event at the target location; wherein the occurrence status information of the dangerous event includes the type of dangerous event, the time of occurrence and the scope of the affected area.
[0037] In the above technical solution, the activity status data of all elements belonging to the hazardous source may overlap and be repeated. In order to reduce the processing workload of the activity status data, the activity status data of all elements are deduplicated based on all behavioral activity types under the activity status data of all elements belonging to the hazardous source, thereby reducing the noise interference of the activity status data. Then, based on the generation time of the activity status data of all elements belonging to the hazardous source from early to late or from late to early, the activity status data that has completed the data deduplication processing is sorted and a corresponding activity status data queue is generated, which can improve the orderliness of the activity status data. The activity status data queue is also input into the corresponding FNN model for summary processing to predict the type, time and scope of the hazardous event at the target location, thereby improving the credibility of the hazardous event prediction.
[0038] Preferably, the image labeling processing module is used to perform neural network labeling processing of elements belonging to the danger source on the original image based on the dangerous event occurrence status information, including: The dangerous event occurrence status information and the personnel flow prediction data within the target location are input into the GAN model for processing to predict the risk level information of the dangerous event; based on the risk level information, the original image is labeled with the location of the elements belonging to the dangerous source and the risk level information of the dangerous event generated; The image output adjustment module is used to determine whether the original image meets the preset picture content label layout conditions based on the result of the neural network labeling processing, so as to adjust the output state of the original image, including: Performing picture information recognition on the original picture that has completed labeling processing to obtain the picture information distribution density of the original picture that has completed labeling processing; if the picture information distribution density is greater than a preset information distribution density threshold, then judging that the original picture meets the preset picture content label layout conditions; otherwise, judging that the original picture does not meet the preset picture content label layout conditions; when the original picture meets the preset picture content label layout conditions, directly outputting the original picture that has completed labeling processing; when the original picture does not meet the preset picture content label layout conditions, re-performing the labeling processing on the original picture until it meets the preset picture content label layout conditions and then directly outputting it.
[0039] In the above technical scheme, if an element inside the target place is exposed to an unexpected dangerous event such as fire or explosion, it will pose a threat to the personnel in the corresponding range inside the target place, thereby generating dangerous events of different high and low risk levels. To this end, the dangerous event occurrence status information and the personnel flow prediction data inside the target place are input into the GAN model for processing to predict the risk level information of the dangerous event; then based on the risk level information, the original image is labeled with the position of the element belonging to the dangerous source and the risk level information of the dangerous event generated, so that the dangerous events that will occur inside the target place can be visualized. In addition, the original image that has been labeled is subjected to picture information recognition to obtain the picture information distribution density of the original image that has been labeled, and the picture information distribution density is compared with a threshold value to determine whether the original image meets the preset picture content label layout conditions, thereby adjusting the output state of the original image, so that the output original image can provide a reliable basis for risk investigation.
[0040] It can be seen from the contents of the above embodiments that the artificial intelligence-based image classification, recognition and risk assessment method and system take several original images of different areas of the target place, and identify the elements of the original images to obtain the element attribute information of the original images, and preliminarily identify the elements inside the target place that may cause unexpected dangers; divide the original image into several image sub-screens corresponding to all elements, and identify the activity status data of the elements through a neural network to determine whether the elements belong to a hazard source, and perform targeted identification of different types of elements; then summarize the activity status data of all elements belonging to the hazard source through a neural network, predict the status information of the occurrence of dangerous events in the target place, and accurately and comprehensively predict the dangerous events caused by the elements; and based on the status information of the occurrence of dangerous events, perform neural network labeling processing on the elements belonging to the hazard source on the original image, so as to realize intelligent and efficient visual recognition of the target place, and ensure that the output original image can provide a reliable basis for risk investigation.
[0041] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims
1. The method of image classification, recognition and risk assessment based on artificial intelligence is characterized by: It includes the following steps: Step S1, photographing a target location to obtain a number of original images of different areas of the target location; identifying screen elements of the original images to obtain element attribute information of the original images; Step S2, based on the element attribute information, performing screen capture processing on the original image to obtain a number of image sub-screens corresponding to all elements one by one; Performing neural network recognition on the image sub-picture to obtain activity status data of the element in the target location; judging whether the element is a danger source inside the target location based on the activity status data; Step S3, summarizing the activity status data of all elements belonging to the hazard source by neural network, and predicting the occurrence status information of the hazard event in the target place; Step S4, based on the dangerous event occurrence status information, performing neural network labeling processing on the original image of elements belonging to the dangerous source; Based on the result of the neural network labeling processing, it is determined whether the original image meets the preset picture content label layout conditions, thereby adjusting the output state of the original image.
2. The method for image classification, recognition and risk assessment based on artificial intelligence as claimed in claim 1, characterized in that: In step S1, the target place is photographed to obtain a number of original pictures of different areas of the target place; the picture elements of the original pictures are identified to obtain the element attribute information of the pictures of the original pictures, including: Performing non-contact detection on the target location to obtain relative position information between different areas inside the target location and the shooting reference point; adjusting the shooting depth of field and shooting field of view of the corresponding area inside the target location based on the relative position information, thereby shooting the corresponding area to obtain an original picture of the corresponding area; After the original image is subjected to image noise reduction filtering and image pixel edge sharpening processing, the original image is input into the CNN model for image element recognition to obtain all elements existing on the original image and their visual attribute information as the element attribute information; wherein the elements include personnel, mechanical equipment, background environment and materials.
3. The method for image classification, recognition and risk assessment based on artificial intelligence as claimed in claim 1, characterized in that: In the step S2, based on the element attribute information, the original image is subjected to screen capture processing to obtain a number of image sub-images corresponding to all elements one by one; the image sub-images are subjected to neural network recognition to obtain activity status data of the elements in the target location; Based on the activity status data, determining whether the element is a danger source inside the target location includes: Based on the visual attribute information of all elements of the original picture, determine the associated picture area of each element in the original picture; based on the boundary of the associated picture area, perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to the element; based on the element types corresponding to all the picture sub-pictures, divide all the picture sub-pictures into a plurality of picture sub-picture sets; wherein all the picture sub-pictures under each picture sub-picture set correspond to the same type of elements; Input all the picture sub-pictures under the picture sub-picture set into the Bayes-FNN model for recognition, and obtain the activity state data of the elements in the target place; wherein the activity state data includes the behavior activity state data of the personnel, mechanical equipment, background environment and materials in the target place; Based on the activity status data, the influence range and duration of the element in the process of performing corresponding behavioral activities inside the target place are obtained; then based on the influence range and duration, it is determined whether the element is a dangerous source inside the target place.
4. The method for image classification, recognition and risk assessment based on artificial intelligence as claimed in claim 1, characterized in that: In step S3, the activity status data of all elements belonging to the hazard source are summarized by a neural network to predict the occurrence status information of the hazard event at the target location, including: Based on all the behavior activity types under the activity status data of all the elements belonging to the hazardous source, the activity status data of all the elements are deduplicated; based on the generation time of the activity status data of all the elements belonging to the hazardous source, the activity status data that has completed the deduplication processing are sorted to generate a corresponding activity status data queue; The activity status data queue is input into the corresponding FNN model for summary processing to predict the occurrence status information of the dangerous event at the target location; wherein the occurrence status information of the dangerous event includes the type of dangerous event, the time of occurrence and the scope of the affected area.
5. The method for image classification, recognition and risk assessment based on artificial intelligence as claimed in claim 1, characterized in that: In the step S4, based on the dangerous event occurrence status information, the original image is subjected to neural network labeling processing of elements belonging to the dangerous source; Based on the result of the neural network labeling processing, it is determined whether the original image meets the preset picture content label layout condition, thereby adjusting the output state of the original image, including: Input the dangerous event occurrence status information and the personnel flow prediction data inside the target place into the GAN model for processing to predict the risk level information of the dangerous event; then based on the risk level information, label the location of the elements belonging to the dangerous source and the risk level information of the dangerous event generated in the original image; Performing picture information recognition on the original picture that has completed labeling processing to obtain the picture information distribution density of the original picture that has completed labeling processing; if the picture information distribution density is greater than a preset information distribution density threshold, then judging that the original picture meets the preset picture content label layout condition; otherwise, judging that the original picture does not meet the preset picture content label layout condition; when the original picture meets the preset picture content label layout condition, directly outputting the original picture that has completed labeling processing; when the original picture does not meet the preset picture content label layout condition, re-performing the labeling processing on the original picture until it meets the preset picture content label layout condition and then directly outputting it.
6. The image classification, recognition and risk assessment system based on artificial intelligence is characterized by: include: An original picture acquisition module is used to photograph a target location and obtain a number of original pictures of different areas of the target location; An element recognition module is used to recognize the elements of the original picture to obtain the element attribute information of the original picture; An original picture segmentation module is used to perform screen capture processing on the original picture based on the element attribute information to obtain a number of picture sub-pictures corresponding to all elements one by one; A hazard source identification module is used to perform neural network recognition on the image sub-picture to obtain activity status data of the element in the target place; based on the activity status data, determine whether the element belongs to a hazard source inside the target place; A dangerous event prediction module is used to aggregate the activity status data of all elements belonging to the dangerous source through a neural network to predict the occurrence status information of the dangerous event in the target place; An image labeling processing module is used to perform neural network labeling processing of elements belonging to the danger source on the original image based on the dangerous event occurrence status information; The image output adjustment module is used to determine whether the original image meets the preset picture content label layout conditions based on the result of the neural network labeling processing, so as to adjust the output state of the original image.
7. The artificial intelligence-based image classification, recognition and risk assessment system according to claim 6, characterized in that: The original picture acquisition module is used to photograph the target location to obtain a number of original pictures of different areas of the target location, including: Performing non-contact detection on the target location to obtain relative position information between different areas inside the target location and the shooting reference point; adjusting the shooting depth of field and shooting field of view of the corresponding area inside the target location based on the relative position information, thereby shooting the corresponding area to obtain an original picture of the corresponding area; The element recognition module is used to perform screen element recognition on the original picture to obtain element attribute information of the original picture, including: After the original image is subjected to image noise reduction filtering and image pixel edge sharpening processing, the original image is input into the CNN model for image element recognition to obtain all elements existing on the original image and their visual attribute information as the element attribute information; wherein the elements include personnel, mechanical equipment, background environment and materials.
8. The artificial intelligence-based image classification, recognition and risk assessment system according to claim 6, characterized in that: The original picture segmentation module is used to perform screen capture processing on the original picture based on the element attribute information to obtain a number of picture sub-pictures corresponding to all elements one by one, including: Based on the visual attribute information of all elements of the original picture, determine the associated picture area of each element in the original picture; based on the boundary of the associated picture area, perform picture interception processing on the original picture to obtain a picture sub-picture corresponding to the element; based on the element types corresponding to all the picture sub-pictures, divide all the picture sub-pictures into a plurality of picture sub-picture sets; wherein all the picture sub-pictures under each picture sub-picture set correspond to the same type of elements; The hazard source identification module is used to perform neural network recognition on the image sub-image to obtain activity status data of the element in the target place; based on the activity status data, determine whether the element belongs to a hazard source inside the target place, including: Input all the picture sub-pictures under the picture sub-picture set into the Bayes-FNN model for recognition, and obtain the activity state data of the elements in the target place; wherein the activity state data includes the behavior activity state data of the personnel, mechanical equipment, background environment and materials in the target place; Based on the activity status data, the influence range and duration of the element in the process of performing corresponding behavioral activities inside the target place are obtained; then based on the influence range and duration, it is determined whether the element is a dangerous source inside the target place.
9. The artificial intelligence-based image classification, recognition and risk assessment system according to claim 6, characterized in that: The dangerous event prediction module is used to aggregate the activity status data of all elements belonging to the dangerous source through a neural network to predict the dangerous event occurrence status information of the target place, including: Based on all the behavior activity types under the activity status data of all the elements belonging to the hazardous source, the activity status data of all the elements are deduplicated; based on the generation time of the activity status data of all the elements belonging to the hazardous source, the activity status data that has completed the deduplication processing are sorted to generate a corresponding activity status data queue; The activity status data queue is input into the corresponding FNN model for summary processing to predict the occurrence status information of the dangerous event at the target location; wherein the occurrence status information of the dangerous event includes the type of dangerous event, the time of occurrence and the scope of the affected area.
10. The artificial intelligence-based image classification, recognition and risk assessment system according to claim 6, characterized in that: The image labeling processing module is used to perform neural network labeling processing of elements belonging to the danger source on the original image based on the dangerous event occurrence status information, including: Input the dangerous event occurrence status information and the personnel flow prediction data inside the target place into the GAN model for processing to predict the risk level information of the dangerous event; then based on the risk level information, label the location of the elements belonging to the dangerous source and the risk level information of the dangerous event generated in the original image; The picture output adjustment module is used to determine whether the original picture meets the preset picture content label layout conditions based on the result of the neural network labeling processing, so as to adjust the output state of the original picture, including: Performing picture information recognition on the original picture that has completed labeling processing to obtain the picture information distribution density of the original picture that has completed labeling processing; if the picture information distribution density is greater than a preset information distribution density threshold, then judging that the original picture meets the preset picture content label layout condition; otherwise, judging that the original picture does not meet the preset picture content label layout condition; when the original picture meets the preset picture content label layout condition, directly outputting the original picture that has completed labeling processing; when the original picture does not meet the preset picture content label layout condition, re-performing the labeling processing on the original picture until it meets the preset picture content label layout condition and then directly outputting it.
