Intelligent automatic inspection vehicle based on AI vision and inspection system thereof

Abstract

The invention relates to the technical field of intelligent inspection, and discloses an intelligent automatic inspection vehicle based on AI vision and an inspection system thereof.The intelligent automatic inspection vehicle comprises a vehicle body, a connecting shaft is rotationally connected to the bottom of the vehicle body, wheels are fixedly connected to the two ends of the connecting shaft, and a plurality of supporting plates are sequentially arranged at the top of the vehicle body; one end of a long rod is fixedly connected to the outer wall of the supporting plate on one side, the other end of the long rod sequentially penetrates through the multiple supporting plates and is fixedly connected with a camera, an image acquisition module and an environment sensor module are carried on the surface of the supporting plate on the other side, and an intelligent inspection system is carried in the vehicle body. The system is used for inspecting the external environment. The image information of the equipment can be quickly acquired and instantly analyzed through the data acquisition and processing module, and information including the appearance state of the equipment and instrument reading can be identified in a short time, so that the inspection time is shortened, and the overall inspection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent inspection technology, and in particular to an AI vision-based intelligent automatic inspection vehicle and an inspection system thereof. Background Art

[0002] In many industrial fields and scenarios, traditional manual inspections have long been one of the main means of equipment and environmental monitoring. However, manual inspections have many significant limitations. Manual inspections rely on inspectors to move around the inspection area on foot or with the help of simple transportation, which is slow and has limited coverage. In large industrial sites such as factory workshops, power substations, and mines, equipment is widely distributed and numerous, and it often takes a lot of time to complete a comprehensive inspection manually.

[0003] In order to overcome the problems of some manual inspections, some semi-automatic inspection equipment have emerged, but these devices also have many shortcomings. Some semi-automatic inspection equipment can only realize simple data acquisition functions, such as single temperature measurement or image capture, but lack effective data analysis and processing capabilities. They have poor adaptability when facing complex environments and changing task requirements, and usually can only perform fixed tasks according to preset programs, and it is difficult to make intelligent decisions and adjustments based on real-time conditions. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides an AI vision-based intelligent automatic inspection vehicle and its inspection system, which solves the problem of poor adaptability of inspection equipment when facing complex environments and changing task requirements.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI vision-based intelligent automatic inspection vehicle, comprising a vehicle body, the bottom of the vehicle body is rotatably connected to a connecting shaft, both ends of the connecting shaft are fixedly connected to wheels, and a plurality of support plates are arranged in sequence on the top of the vehicle body, an outer wall of the support plate on one side is fixedly connected to one end of a long rod, the other end of the long rod passes through a plurality of support plates in sequence and is fixedly connected to a camera, and an image acquisition module and an environmental sensor module are respectively mounted on the surface of the support plate on the other side, and an intelligent inspection system is mounted inside the vehicle body for inspecting the external environment.

[0006] Preferably, the vehicle body is equipped with a SLAM function, which uses laser radar, visual camera and inertial measurement unit to perform multi-sensor fusion environmental perception.

[0007] In addition, the present invention also provides an intelligent automatic inspection system based on AI vision, comprising:

[0008] A data acquisition and processing module is used to receive and process the data collected by the image acquisition module and the environmental sensor module;

[0009] Map construction module, which builds and updates the map in real time;

[0010] Intelligent analysis and decision-making module, which receives processed data for analysis and makes decisions;

[0011] The communication and control module is used for communication between internal modules and interactive control with the outside.

[0012] Preferably, the environmental sensor includes a gas sensor and a temperature sensor, and the gas sensor concentration calibration formula is:

[0013]

[0014] Among them, C gas is the measured value of gas concentration, S gas is the sensor output signal value, a i is the calibration factor;

[0015] The temperature sensor temperature calibration formula is:

[0016] ΔT=T s -T m +b

[0017] Among them, T s is the standard thermometer measurement value, T m is the measured value of the temperature sensor to be calibrated, and b is the compensation value after calibration.

[0018] Preferably, the data acquisition and processing module includes:

[0019] The data preprocessing unit performs cropping, scaling, and normalization on the image data, performs signal conditioning and filtering on the environmental sensor data, and uses a sliding window mean filtering algorithm to smooth the sensor data. The formula is:

[0020]

[0021] in, is the value of the nth data point after filtering, D i is the value of the original data point, m is the sliding window size;

[0022] The data fusion unit fuses the preprocessed image data with the sensor data and combines it with the visual data to perform status updates.

[0023] Preferably, the map construction module includes:

[0024] The feature extraction unit extracts the feature points and line segment features in the image by using a method based on the combination of feature points and line features;

[0025] The construction unit uses the lidar data for fusion and construction of the map;

[0026] Update unit, uses iterative closest point algorithm to update point cloud registration.

[0027] Preferably, the intelligent analysis and decision-making module includes:

[0028] AI visual recognition module, which uses a deep learning convolutional neural network model for target detection and classification;

[0029] Path analysis unit, to plan the inspection vehicle's driving path;

[0030] The alarm unit, when an abnormal situation is detected, sounds an alarm through the sound and light alarm device.

[0031] Preferably, the communication and control module includes:

[0032] Data transmission unit, which transmits environmental collection data, map construction data and decision data;

[0033] The data compression unit establishes a sliding window, matches the data to be compressed with the existing data in the window, and compresses the collected data;

[0034] The data encryption unit generates round keys through a key expansion algorithm, and converts plaintext data into ciphertext data through multiple rounds of byte replacement, row shift, column mixing and round key addition operations.

[0035] Preferably, during the inspection task execution, a dynamic threshold detection algorithm is used for the gas concentration data, and the detection threshold is dynamically adjusted according to historical data and real-time environmental changes; assuming that the current gas concentration data sequence is {C i}, its moving average formula is:

[0036]

[0037] Detection Threshold:

[0038]

[0039] Among them, σ is the standard deviation. When the gas concentration collected in real time exceeds the threshold, it is judged as an abnormal situation. Among them, m is a coefficient determined according to safety standards and environmental factors.

[0040] Preferably, during the process of path planning, the path analysis unit adopts a path planning algorithm based on a cost map to convert obstacle information, dangerous area information, etc. into cost values ​​in the cost map. When encountering temporary obstacles and environmental changes, the local path is replanned using a local path planning algorithm based on the artificial potential field method.

[0041] Working principle: During the data acquisition phase, the camera, image acquisition module and environmental sensor module on the vehicle body start working. The camera captures image information of the surrounding environment, the laser radar, inertial measurement unit and odometer obtain information such as the position, posture and distance of the vehicle body from different angles, and the gas sensor and temperature sensor in the environmental sensor detect the concentration and temperature of gas components in the environment respectively. In the data acquisition and processing module, the data preprocessing unit first performs processing such as cropping, scaling, and normalization on the image data to improve the image quality, and performs signal conditioning and filtering on the sensor data. The data fusion unit fuses the processed image data and sensor data, uses the Kalman filter algorithm to predict and correct the sensor data, and establishes a dynamic threshold model based on environmental factors to adjust the gas concentration threshold, and combines time series analysis and image information to comprehensively judge abnormal situations.

[0042] In the map construction module, the feature extraction unit extracts feature points and line segment features from the image and fuses them with the lidar data. The initial map is created by obtaining environmental distance, posture, and movement distance information through sensors such as lidar. The iterative closest point algorithm is used for scanning and matching to find the best position and posture of the point cloud in the map. At the same time, closed-loop detection is performed. Multiple inspection vehicles can perform map fusion and use the iterative closest point algorithm to update the map.

[0043] In the intelligent analysis and decision-making module, the AI ​​visual recognition module uses deep learning convolutional neural networks for target detection and classification. The path analysis unit uses a path planning algorithm based on a cost map to plan the path globally, and calculates the minimum cost path based on the grid unit cost weight. When encountering temporary obstacles, the artificial potential field method is used to replan the local path, and the driving direction is determined by calculating the direction of the resultant force through the gravitational field function and the repulsive field function, and the speed and steering angle are determined in combination with the vehicle kinematic model.

[0044] In the communication and control module, the data transmission unit selects the appropriate transmission protocol according to the data type. The UDP protocol is used to transmit environmental collection data to ensure real-time performance, and the TCP protocol is used to ensure accuracy of map construction and decision-making data. The data compression unit uses a sliding window algorithm to adjust the window size according to the characteristics of the data, expands the window for highly repetitive data to improve the compression ratio, and reduces the window for fast-changing data to improve the search speed. The data encryption unit generates round keys through a key expansion algorithm, and converts plaintext into ciphertext through multiple rounds of encryption operations to ensure secure data transmission.

[0045] The present invention provides an intelligent automatic inspection vehicle based on AI vision and its inspection system. It has the following beneficial effects:

[0046] 1. The present invention can quickly acquire the image information of the equipment and conduct instant analysis through the data acquisition and processing module, and can identify information including the appearance status of the equipment and instrument readings in a short time, thereby shortening the inspection time and improving the overall inspection efficiency.

[0047] 2. The present invention has a SLAM function. Through the fusion of multiple sensors, a high-precision three-dimensional map is constructed in real time under the improved GMapping algorithm. It can adapt to various scenarios including construction site monitoring, mine safety inspections, industrial park logistics, emergency rescue and environmental monitoring.

[0048] 3. The environmental sensor in the present invention is optimized for installation according to different gas characteristics and distribution patterns, and adopts a variety of calibration methods to ensure data accuracy. It can work stably regardless of high temperature, high humidity, dust or severe electromagnetic interference environment, further improving the adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A three-dimensional diagram of the inspection vehicle of the present invention;

[0050] Figure 2 It is a schematic diagram of the inspection system in the present invention;

[0051] Figure 3 It is a schematic diagram of the data acquisition and processing module in the present invention;

[0052] Figure 4 A schematic diagram of a map construction module in the present invention;

[0053] Figure 5 Schematic diagram of the intelligent analysis and decision-making module in the present invention;

[0054] Figure 6 It is a schematic diagram of the communication and control module in the present invention.

[0055] Among them, 1. Car body; 2. Connecting shaft; 3. Wheels; 4. Support plate; 5. Long pole; 6. Camera. DETAILED DESCRIPTION

[0056] 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.

[0057] Please see attached Figure 1 -Attached Figure 6The embodiment of the present invention provides an AI vision-based intelligent automatic inspection vehicle, including a vehicle body 1, the bottom of the vehicle body 1 is rotatably connected to a connecting shaft 2, both ends of the connecting shaft 2 are fixedly connected to wheels 3, a plurality of support plates 4 are sequentially arranged on the top of the vehicle body 1, an outer wall of the support plate 4 on one side is fixedly connected to one end of a long rod 5, the other end of the long rod 5 passes through the plurality of support plates 4 in sequence and is fixedly connected to a camera 6, the surfaces of the support plates 4 on the other side are respectively equipped with an image acquisition module and an environmental sensor module, and the interior of the vehicle body 1 is equipped with an intelligent inspection system for inspecting the external environment.

[0058] The vehicle body 1 is equipped with a SLAM function, which uses laser radar, visual camera and inertial measurement unit to perform multi-sensor fusion environmental perception;

[0059] Specifically, the vehicle body 1 is used as a carrier, and movement is achieved by connecting the shaft 2 and the wheel 3. The top support plate 4 is equipped with a long pole 5, a camera 6, an image acquisition module and an environmental sensor module, and is equipped with an intelligent inspection system and a SLAM function with multi-sensor fusion, so as to achieve comprehensive perception and inspection of the environment, which can be achieved: monitoring the progress and quality of construction, such as comparing images to determine the progress of construction, and detecting whether the construction process is up to standard; it can also monitor personnel safety and equipment operation status, including personnel identity recognition, helmet wearing detection and equipment failure hidden danger inspection; conduct safety inspections in mining areas, such as detecting the risk of mine collapse and illegal operation of transportation equipment; at the same time, it can monitor the environment and resource conditions, such as air quality, dust concentration and ore reserve estimation; manage logistics vehicles and goods, such as identifying vehicle information, optimizing driving routes, and detecting cargo packaging and stacking; it can also monitor park facilities and environment, such as monitoring infrastructure and vegetation growth.

[0060] In addition, the present invention also provides an intelligent automatic inspection system based on AI vision, comprising:

[0061] A data acquisition and processing module is used to receive and process the data collected by the image acquisition module and the environmental sensor module;

[0062] Map construction module, which builds and updates the map in real time;

[0063] Intelligent analysis and decision-making module, which receives processed data for analysis and makes decisions;

[0064] The communication and control module is used for communication between internal modules and interactive control with the outside.

[0065] Environmental sensors include gas sensors and temperature sensors. The gas sensor concentration calibration formula is:

[0066]

[0067] Among them, Cgas is the measured value of gas concentration, S gas is the sensor output signal value, a i is the calibration factor;

[0068] The temperature sensor temperature calibration formula is:

[0069] ΔT=T s -T m +b

[0070] Among them, T s is the standard thermometer measurement value, T m is the measured value of the temperature sensor to be calibrated, and b is the compensation value after calibration;

[0071] During the inspection task execution, a dynamic threshold detection algorithm is used for gas concentration data, and the detection threshold is dynamically adjusted according to historical data and real-time environmental changes; assuming that the current gas concentration data sequence is {C i}, its moving average formula is:

[0072]

[0073] Detection Threshold:

[0074]

[0075] Wherein, σ is the standard deviation. When the real-time collected gas concentration exceeds the threshold, it is judged as an abnormal situation. Wherein, m is a coefficient determined according to safety standards and environmental factors;

[0076] Specifically, the gas sensor uses a combination of electrochemical sensors and infrared sensors. Electrochemical sensors are highly sensitive to common harmful gases such as hydrogen sulfide and carbon monoxide, and can quickly detect low-concentration gas leaks. Infrared sensors are used to detect gases such as carbon dioxide. According to the diffusion characteristics and density distribution of gases, an electrochemical sensor array is installed at the bottom of the inspection vehicle near the ground to detect heavy gases that may be deposited near the ground; infrared sensors are installed at a higher position on the roof to monitor the distribution of light gases in the air. The temperature sensor uses a platinum resistance temperature sensor, which is attached to the key heating parts of the equipment through thermal conductive silicone. The non-contact infrared temperature sensor uses focal plane array technology and can quickly scan the temperature distribution of a large area. During the sensor calibration process, a standard gas generator is used to generate a gas sample of precise concentration. After the gas sensor is placed in the sample environment for a period of time, the sensor output value is recorded. By comparing the standard concentration value and the sensor output value, the coefficient of the calibration curve is calculated using the least squares method to establish an accurate relationship between the gas concentration and the sensor output signal.

[0077] The data acquisition and processing modules include:

[0078] The data preprocessing unit performs cropping, scaling, and normalization on the image data, performs signal conditioning and filtering on the environmental sensor data, and uses a sliding window mean filtering algorithm to smooth the sensor data. The formula is:

[0079]

[0080] in, is the value of the nth data point after filtering, D i is the value of the original data point, m is the sliding window size;

[0081] The data fusion unit fuses the pre-processed image data with the sensor data and combines the visual data for status update;

[0082] Specifically, firstly, the sensor data is synchronized in time and space to ensure the consistency of the data in time and space dimensions. The sensor data is filtered using the Kalman filter algorithm to predict the sensor value at the next moment and make corrections based on the actual measured value. For the gas concentration data, a dynamic threshold model is established to dynamically adjust the threshold based on factors such as ambient temperature, humidity, and equipment operation status. For example, when the ambient temperature rises, the volatilization rate of some gases accelerates, and the warning threshold of the gas concentration is correspondingly increased. Combined with the time series analysis method, the ARI MA model is used to predict the change trend of the gas concentration data, and the equipment surface temperature distribution information obtained by image analysis is fused with the temperature sensor measurement data, and more accurate temperature field information is obtained through weighted averaging and other methods. When an abnormal change in gas concentration is detected, a comprehensive analysis and judgment is performed in combination with the operating status of the equipment in the image, the activities of the surrounding personnel, and other information to determine the cause and location of the abnormality. If the image shows that the equipment is smoking and the gas sensor detects an increase in the concentration of harmful gases, it is judged that the equipment has failed and caused a gas leak.

[0083] Map building blocks include:

[0084] The feature extraction unit extracts the feature points and line segment features in the image by using a method based on the combination of feature points and line features;

[0085] The construction unit uses the lidar data for fusion and construction of the map;

[0086] Update unit, which uses iterative closest point algorithm to update point cloud registration;

[0087] Specifically, for the extraction of image feature points, scale-invariant feature transformation or accelerated robust feature algorithm is used to detect key points in the image and calculate its feature descriptor. For the extraction of line segment features, Hough transform is used to extract line segment information from the image, and these image features are matched and fused with the point cloud data of the lidar. The corresponding feature points and line segments are found through the feature matching algorithm to achieve the fusion of multi-sensor data and improve the accuracy and completeness of map construction.

[0088] In the process of map construction, the environmental data is collected through the assembly of sensor equipment such as laser radar, inertial measurement unit, and odometer: the laser radar is used to obtain the distance information of the environment, the IMU provides attitude information, and the odometer assists in measuring the moving distance. According to the initial position and attitude of the inspection vehicle, an initial blank octree map is created, and the laser radar point cloud at the current moment is scanned and matched with the map. The best matching position and attitude of the laser radar point cloud in the map are found through the iterative nearest point algorithm. In the process of construction, it is detected in real time whether the robot returns to the position that has been visited before, that is, the closed loop. The closed loop detection can be achieved by comparing the similarity between the current laser radar point cloud and the historical point cloud. When multiple inspection vehicles build maps at the same time, map fusion can be performed to align and merge different maps to improve the integrity and accuracy of the map. After the map is built, the constructed map can be applied to the navigation and positioning of the inspection vehicle. The inspection vehicle can perform path planning, obstacle avoidance and other operations based on the map information, and use the map to locate itself and continuously update the position and attitude estimation;

[0089] During the map update process, the iterative closest point algorithm is used for point cloud registration. The formula is:

[0090]

[0091] Among them, P i is a point in the target point cloud, q i is a point in the source point cloud, R is the rotation matrix, and t is the translation vector. By continuously iteratively optimizing R and t, accurate point cloud registration and map update are achieved. Each time new sensor data is obtained, it is registered and fused with the existing map, and the map information is updated to reflect changes in the environment.

[0092] Intelligent analysis and decision-making modules include:

[0093] AI visual recognition module, which uses a deep learning convolutional neural network model for target detection and classification;

[0094] Path analysis unit, to plan the inspection vehicle's driving path;

[0095] An alarm unit, which sounds an alarm through an audible and visual alarm device when an abnormal situation is detected;

[0096] In the process of path planning, the path analysis unit uses a path planning algorithm based on the cost map to convert obstacle information, dangerous area information, etc. into cost values ​​in the cost map. When encountering temporary obstacles and environmental changes, the local path planning algorithm based on the artificial potential field method is used to replan the local path.

[0097] Specifically, a map model of the inspection area is constructed based on the grid method, and the map is divided into grid units of the same size. In the global path planning stage, different cost weights are assigned to each grid unit according to the inspection task requirements and equipment distribution. For example, the grid cost is higher near dangerous areas or obstacles, while the cost is lower in open areas without obstacles. By calculating the cumulative cost from the starting grid to the target grid, the path with the minimum cost is found as the global inspection path. During the driving process of the inspection vehicle, lidar and cameras are used to perceive changes in the surrounding environment in real time. When encountering temporary obstacles or environmental changes, the local path is replanned using a local path planning algorithm based on the artificial potential field method. Gravitational field function:

[0098]

[0099] Repulsive field function:

[0100]

[0101] The total force on the vehicle:

[0102]

[0103] Among them, k att is the gravitational coefficient, which is determined according to the distance and importance of the target point; x is the current position coordinate of the vehicle, x goal is the target position coordinate, k rep is the repulsion coefficient, which is determined according to the danger level and distance of the obstacle; x obsis the obstacle position coordinate, and ρ0 is the radius of the obstacle influence range. In the process of cost map construction, each grid in the map is assigned a corresponding cost value according to the obstacle information detected by the sensor, the dangerous area detected by the environmental sensor (such as high temperature, high concentration of harmful gas area), etc. In path planning, the path with the minimum cost is searched as the driving path of the inspection vehicle. When encountering a new obstacle, the inspection vehicle is regarded as a particle, the target point generates gravity, and the obstacle generates repulsion. The magnitude of gravity is proportional to the distance from the inspection vehicle to the target point. The gravity and repulsion are calculated in real time by the artificial potential field method. The magnitude of repulsion is inversely proportional to the distance from the inspection vehicle to the obstacle. The driving direction of the inspection vehicle is determined by calculating the direction of the combined force of gravity and repulsion. At the same time, combined with the kinematic model of the vehicle, according to the current speed and acceleration constraints, the appropriate driving speed and steering angle are calculated to ensure that the inspection vehicle can safely and quickly avoid obstacles and continue to drive along the inspection path.

[0104] The communication and control module includes:

[0105] Data transmission unit, which transmits environmental collection data, map construction data and decision data;

[0106] The data compression unit establishes a sliding window, matches the data to be compressed with the existing data in the window, and compresses the collected data;

[0107] The data encryption unit generates round keys through a key expansion algorithm, and converts plaintext data into ciphertext data through multiple rounds of byte replacement, row shift, column mixing and round key addition operations;

[0108] Specifically, the use of UDP-based optimization protocols can quickly send out the raw data collected by lidar, cameras and various sensors, reducing the delay in data transmission so that subsequent modules can process them in time. For information such as map construction data and decision data that have strict accuracy requirements, the TCP protocol or its derived reliable transmission protocol is used to ensure that the data reaches the destination intact and without error by establishing connections, confirming responses and retransmission mechanisms. The sliding window algorithm of the data compression unit is the core of achieving efficient data compression. When establishing a sliding window, the search efficiency and matching accuracy of the window are improved by optimizing the algorithm. An adaptive window size adjustment strategy is used to dynamically change the size of the window according to the characteristics and repetitive patterns of the data. For fragments with high data repetitiveness, the window is appropriately expanded to increase the matching length, thereby achieving a higher compression ratio; for parts with fast data changes and low repetitiveness, the window is reduced to increase the search speed and reduce unnecessary computing overhead. The round key generated by the key expansion algorithm of the data encryption unit is the key to ensuring data security. Through a strict key management mechanism, the security of the key generation, storage, distribution and update process is ensured. A high-intensity random number generator is used to generate the initial key, and the key is stored in combination with a hardware encryption module to prevent the key from being stolen or tampered with. In the key distribution process, encryption channels and digital certificate technology are used to ensure that only authorized recipients can obtain the key. At the same time, round keys are updated regularly to deal with possible key leakage risks and improve the security and reliability of data encryption.

[0109] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI vision-based intelligent automatic inspection vehicle, comprising a vehicle body (1), characterized in that: The bottom of the vehicle body (1) is rotatably connected to a connecting shaft (2), and both ends of the connecting shaft (2) are fixedly connected to wheels (3). The top of the vehicle body (1) is provided with a plurality of support plates (4) in sequence, and the outer wall of the support plate (4) on one side is fixedly connected to one end of a long rod (5), and the other end of the long rod (5) passes through the plurality of support plates (4) in sequence and is fixedly connected to a camera (6). The surfaces of the support plates (4) on the other side are respectively equipped with an image acquisition module and an environmental sensor module. The interior of the vehicle body (1) is equipped with an intelligent inspection system for inspecting the external environment.

2. The AI ​​vision-based intelligent automatic inspection vehicle according to claim 1 is characterized in that: The vehicle body (1) is equipped with a SLAM function, which uses a laser radar, a visual camera and an inertial measurement unit to perform multi-sensor fusion environmental perception.

3. An intelligent automatic inspection system based on AI vision, characterized in that: An AI vision-based intelligent automatic inspection vehicle applied to claims 1-2, comprising: A data acquisition and processing module is used to receive and process the data collected by the image acquisition module and the environmental sensor module; Map construction module, which builds and updates the map in real time; Intelligent analysis and decision-making module, which receives processed data for analysis and makes decisions; The communication and control module is used for communication between internal modules and interactive control with the outside.

4. According to claim 3, an intelligent automatic inspection system based on AI vision is characterized in that: The environmental sensor includes a gas sensor and a temperature sensor, and the gas sensor concentration calibration formula is: Among them, C gas is the measured value of gas concentration, S gas is the sensor output signal value, a i is the calibration factor; The temperature sensor temperature calibration formula is: ΔT=T s -T m +b Among them, T s is the standard thermometer measurement value, T m is the measured value of the temperature sensor to be calibrated, and b is the compensation value after calibration.

5. The intelligent automatic inspection system based on AI vision according to claim 3 is characterized in that: The data acquisition and processing module includes: The data preprocessing unit performs cropping, scaling, and normalization on the image data, performs signal conditioning and filtering on the environmental sensor data, and uses a sliding window mean filtering algorithm to smooth the sensor data. The formula is: in, is the value of the nth data point after filtering, D i is the value of the original data point, m is the sliding window size; The data fusion unit fuses the preprocessed image data with the sensor data and combines it with the visual data to perform status updates.

6. The intelligent automatic inspection system based on AI vision according to claim 3 is characterized in that: The map construction module includes: The feature extraction unit extracts the feature points and line segment features in the image by using a method based on the combination of feature points and line features; The construction unit uses the lidar data for fusion and construction of the map; Update unit, uses iterative closest point algorithm to update point cloud registration.

7. The intelligent automatic inspection system based on AI vision according to claim 3 is characterized in that: The intelligent analysis and decision-making module includes: AI visual recognition module, which uses a deep learning convolutional neural network model for target detection and classification; Path analysis unit, to plan the inspection vehicle's driving path; The alarm unit, when an abnormal situation is detected, sounds an alarm through the sound and light alarm device.

8. The intelligent automatic inspection system based on AI vision according to claim 3 is characterized in that: The communication and control module comprises: Data transmission unit, which transmits environmental collection data, map construction data and decision data; The data compression unit establishes a sliding window, matches the data to be compressed with the existing data in the window, and compresses the collected data; The data encryption unit generates round keys through a key expansion algorithm, and converts plaintext data into ciphertext data through multiple rounds of byte replacement, row shift, column mixing and round key addition operations.

9. The intelligent automatic inspection system based on AI vision according to claim 3 is characterized in that: During the inspection task execution, a dynamic threshold detection algorithm is used for gas concentration data, and the detection threshold is dynamically adjusted according to historical data and real-time environmental changes; assuming that the current gas concentration data sequence is {C i }, its moving average formula is: Detection Threshold: Among them, σ is the standard deviation. When the gas concentration collected in real time exceeds the threshold, it is judged as an abnormal situation. Among them, m is a coefficient determined according to safety standards and environmental factors.

10. The intelligent automatic inspection system based on AI vision according to claim 7, characterized in that: During the path planning process, the path analysis unit adopts a path planning algorithm based on a cost map to convert obstacle information, dangerous area information, etc. into cost values ​​in the cost map. When encountering temporary obstacles and environmental changes, the local path is replanned using a local path planning algorithm based on an artificial potential field method.

Images