Vision-based shovel loading efficiency analysis and truck full load identification system

Abstract

The application provides a visual-based excavator loading efficiency analysis and truck full-load identification system, and relates to the technical field of intelligent monitoring of engineering machinery operation. The system comprises: a camera for collecting real-time video streams of an excavator operation site; an edge computing device for receiving the real-time video streams; an AI classification module for performing four-classification identification on the real-time video streams by using a trained deep convolutional neural network image recognition model; a state machine logic module for state discrimination and state conversion according to the loading state classification results, and for accumulating the number of loading times, identifying the full-load event and recording the loading duration of each truck; and a data recording and analysis module for saving the number of loading times, the full-load event and the loading duration of each truck to generate loading efficiency analysis data. In this way, the installation complexity and cost are reduced, the identification precision and detection reliability are improved, the loading efficiency data can be automatically and comprehensively counted, and the data can be obtained in a timely manner.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of construction machinery operation, and in particular to a visual-based excavator loading efficiency analysis and truck full-load identification system. BACKGROUND

[0002] In earthwork construction, open-pit mining operation and other scenarios, excavators usually need to load materials into transport trucks for transfer. This loading process is an important link affecting construction efficiency and capacity. At present, the monitoring and statistics of the loading process of excavators still mostly rely on manual observation or simple sensor means. For example, the number of loading times, the number of loaded buckets, whether each truck is fully loaded, and the time spent for loading each truck, etc. are usually obtained by manual counting or experience estimation. This manual statistical method has many shortcomings: first, it requires special personnel to record on site, increasing labor costs; second, manual counting is prone to omissions or errors, and its accuracy is difficult to guarantee, which may lead to incorrect loading quantity statistics and result in uneven or missed loading; third, it is difficult for managers to discover and improve inefficient links in the loading operation in a timely manner.

[0003] There are also some improvements in the prior art for loading monitoring by means of sensors or simple automation, such as installing weight sensors on transport vehicles to detect full load, or installing angle sensors, vibration sensors and other auxiliary judgment devices on excavators to assist in judging the action of the bucket. However, the use of sensors or simple automation for loading detection often has problems such as high cost, complex installation, poor environmental adaptability, etc. In the case of complex environment, serious dust and light interference in open-pit mines or construction sites, the accuracy and reliability of the sensors may be affected, resulting in low recognition accuracy and low detection reliability, and the inability to obtain data in a timely manner. At the same time, the current excavator working condition monitoring system has a single function and can only provide limited early warning or counting, and cannot automatically and comprehensively count the loading efficiency data. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a visual-based excavator loading efficiency analysis and truck full-load identification system, which solves the problems of high cost, complex installation, low recognition accuracy, low detection reliability, inability to obtain data in a timely manner and inability to automatically and comprehensively count the loading efficiency data in the prior art.

[0005] To solve the above technical problems, the embodiments of the present application provide the following technical solutions:

[0006] The present application provides a visual-based excavator loading efficiency analysis and truck full-load identification system, comprising:

[0007] A camera is installed on the excavator for collecting real-time video streams of the excavator operation site.

[0008] An edge computing device connected with the camera, configured to receive the real-time video stream;

[0009] An AI classification module deployed on the edge computing device, configured to input the real-time video stream into a trained deep convolutional neural network image recognition model, to classify the real-time video stream by using the trained deep convolutional neural network image recognition model, and output a loading state classification result, the loading state classification result including a first shovel category, a loading-in-progress category, a last shovel category, and a preparation state category;

[0010] A state machine logic module connected with the AI classification module, configured to perform state discrimination and state conversion according to the loading state classification result, and accumulate the loading times, identify the loading full load events, and record the loading duration of each vehicle;

[0011] A data recording and analysis module connected with the state machine logic module, configured to save the loading times, the loading full load events, and the loading duration of each vehicle, to generate loading efficiency analysis data, and the loading efficiency analysis data is used to analyze the loading efficiency of the excavator, optimize the vehicle scheduling, and optimize the loading coordination.

[0012] In some embodiments, the trained deep convolutional neural network image recognition model includes a trained ResNet34 model, the trained ResNet34 model is obtained by training a ResNet34 model using a plurality of excavator operation video data and annotations with each frame state, the annotations with each frame state including annotations of the first shovel category, the loading-in-progress category, the last shovel category, and the preparation state category.

[0013] In some embodiments, the state machine logic module is further configured to set a plurality of state nodes of the excavator loading process, the plurality of state nodes including an idle state, a start verification state, a loading-in-progress state, and a cooling state.

[0014] In some embodiments, the state machine logic module is specifically configured to determine that the state machine is in the idle state when the loading state classification result is the preparation state category; when detecting a continuous first shovel category in the idle state, the state machine is switched from the idle state to the start verification state; when detecting a continuous loading-in-progress category in the start verification state, the state machine is switched from the start verification state to the loading-in-progress state, and the loading times are accumulated and the start time of the loading of each vehicle is recorded; when detecting a continuous last shovel category in the loading-in-progress state, the state machine is switched from the loading-in-progress state to the cooling state, and the end time of the loading of each vehicle is recorded, the loading duration of each vehicle is determined based on the start time of the loading of each vehicle and the end time of the loading of each vehicle, and the loading full load event is identified, the state machine returns to the idle state and waits for the next loading period until all loading periods are completed.

[0015] In some embodiments, the vision-based excavator loading efficiency analysis and truck full load identification system further comprises an uploading module;

[0016] The uploading module is connected with the data recording and analysis module, and is configured to transmit the loading efficiency analysis data to a dispatch center in real time.

[0017] In some embodiments, the vision-based excavator loading efficiency analysis and truck full load identification system further comprises:

[0018] The display terminal interface is connected with the data recording and analysis module, and is configured to receive an audible and visual or status signal generated when the truck is full, and send the audible and visual or status signal to the cab display to remind that the current vehicle is full.

[0019] In some embodiments, the data recording and analysis module is further configured to save the number of times of digging, the vehicle identification, the loading sequence number, the start time of loading for each vehicle, and the end time of loading for each vehicle.

[0020] In some embodiments, the edge computing device is an industrial computer, an embedded AI acceleration terminal, or a software and hardware integrated device.

[0021] In some embodiments, the edge computing device is provided with an artificial intelligence hardware acceleration unit, and the artificial intelligence hardware acceleration unit is a graphics processing unit or a neural network acceleration chip.

[0022] Compared with the prior art, the visual-based excavator loading efficiency analysis and truck full load identification system provided by the application comprises: a camera installed on the excavator for collecting real-time video stream of the excavator work site; an edge computing device connected with the camera for receiving the real-time video stream; an AI classification module deployed on the edge computing device for inputting the real-time video stream into a trained deep convolutional neural network image recognition model to classify the real-time video stream by using the trained deep convolutional neural network image recognition model, outputting a loading state classification result, and the loading state classification result comprising a first shovel category, a loading-in-progress category, a last shovel category and a preparation state category; a state machine logic module connected with the AI classification module for state discrimination and state conversion according to the loading state classification result, and accumulating the number of loading trucks, identifying the loading full load event and recording the loading duration of each truck; and a data recording and analysis module connected with the state machine logic module for saving the number of loading trucks, the loading full load event and the loading duration of each truck to generate loading efficiency analysis data, and the loading efficiency analysis data being used for analyzing the loading efficiency of the excavator, optimizing vehicle scheduling and loading coordination. In this way, the real-time video stream of the excavator work site is collected based on the camera to realize full-process monitoring of the loading state of the excavator, and the loading counting and full load identification can be completed without the aid of additional hardware such as weight sensors and position sensors, thereby reducing the installation complexity and cost; the trained deep convolutional neural network image recognition model can accurately distinguish the first shovel category, the loading-in-progress category, the last shovel category and the preparation state category, thereby reducing misjudgment and missing judgment, improving the recognition accuracy and detection reliability; the state machine logic module can discriminate and convert states according to the loading state classification result, accumulate the number of loading trucks, identify the loading full load event and record the loading duration of each truck, and automatically and comprehensively count the loading efficiency data; the AI classification module, the state machine logic module and the data recording and analysis module are real-time running on the edge computing device of the excavator, have low dependence on the network and can obtain data in time. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and other objects, features and advantages of the exemplary embodiments of the present application will be more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which several embodiments of the present application are shown by way of example, and wherein like reference numerals refer to like elements throughout. In the drawings:

[0024] Figure 1 A structure of a visual-based excavator loading efficiency analysis and truck full load identification system is schematically shown Figure 1 ;

[0025] Figure 2 A structure of a visual-based excavator loading efficiency analysis and truck full load identification system is schematically shownFigure 2 ;

[0026] Figure 3 A workflow diagram of the state machine logic module of the excavator is schematically shown. DETAILED DESCRIPTION

[0027] The embodiments of the present application will be further described in detail with reference to the drawings and examples. The detailed description and drawings of the following examples are intended to illustrate the principles of the present application by way of example only, and should not be used to limit the scope of the present application, which can be realized in many different forms, not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

[0028] A visual-based excavator loading efficiency analysis and truck full load identification system in an embodiment of the present application will be described in detail below.

[0029] Referring to Figure 1 and Figure 2 , Figure 1 a structural schematic diagram of the visual-based excavator loading efficiency analysis and truck full load identification system is schematically shown Figure 1 , Figure 2 a structural schematic diagram of the visual-based excavator loading efficiency analysis and truck full load identification system is schematically shown Figure 2 , the embodiment of the present application proposes a visual-based excavator loading efficiency analysis and truck full load identification system, comprising:

[0030] The camera 101 is installed on the excavator and is used to collect real-time video streams of the working site of the excavator;

[0031] The edge computing device 102 is connected with the camera and is used to receive the real-time video streams;

[0032] The AI classification module 103 is deployed on the edge computing device and is used to input the real-time video streams into a trained deep convolutional neural network image recognition model, so as to classify the real-time video streams by using the trained deep convolutional neural network image recognition model, and output a loading state classification result, which includes a first shovel category, a loading-in-progress category, a last shovel category and a preparation state category;

[0033] The state machine logic module 104 is connected with the AI classification module and is used to make state discrimination and state conversion according to the loading state classification result, and accumulate the number of loading trucks, identify the full load event of the loading truck and record the loading time of each truck;

[0034] The data recording and analysis module 105 is connected with the state machine logic module, and is used to save the loading times, loading full events and loading time of each vehicle, so as to generate loading efficiency analysis data, which is used to analyze the loading efficiency of the excavator, optimize vehicle scheduling and loading cooperation.

[0035] Specifically, the camera and the edge computing device can be installed on the arm of the excavator or the top of the cab. The AI classification module, the state machine logic module and the data recording and analysis module are all deployed on the edge computing device. The camera is aimed at the working area of the excavator (such as the position where the bucket interacts with the transport truck), and the camera is used to obtain real-time video stream of the whole loading process. The high-frame-rate real-time video stream collected by the camera is transmitted to the edge computing device for real-time processing through wired or wireless mode.

[0036] Based on single camera visual perception, the loading state monitoring can be realized without installing additional hardware such as weighing sensors and angle sensors, and the key events and states in the loading process can be accurately identified.

[0037] In the embodiment, the edge computing device 102 is an industrial computer, an embedded AI acceleration terminal or a software and hardware integrated device. The edge computing device 102 is provided with an artificial intelligence hardware acceleration unit, which is a graphics processing unit (GPU) or a neural network processing unit (NPU). The GPU and NPU can accelerate the inference calculation of the trained deep convolutional neural network image recognition model, so as to ensure the real-time classification processing of the real-time video stream and the low-delay operation of the system.

[0038] In the embodiment, the trained deep convolutional neural network image recognition model includes a trained ResNet34 model, which is obtained by training the ResNet34 model using a plurality of excavator working video data and annotations with each frame state, including annotations of the first shovel category, annotations of the loading category, annotations of the last shovel category and annotations of the preparation state category.

[0039] Specifically, the AI classification module running on the edge computing device adopts a trained deep convolutional neural network image recognition model to realize intelligent recognition of real-time video stream pictures. The trained deep convolutional neural network image recognition model in the application can be a trained ResNet34 model. There can be multiple trained deep convolutional neural network image recognition models, which are not specifically limited herein. The trained ResNet34 model is selected as the image classification core. The input of the trained ResNet34 model is a single frame image or a sequence frame in a real-time video stream, and the output of the trained ResNet34 model is one of the first shovel category, the loading category, the last shovel category, and the preparation state category in the loading state classification result.

[0040] In order to improve the classification accuracy, a large amount of excavator operation video data is used in the training of the ResNet34 model, and supervised learning is performed in combination with the annotation of each frame, i.e., the annotation of each frame state (annotation of the first shovel category, annotation of the loading category, annotation of the last shovel category, and annotation of the preparation state category). The trained ResNet34 model can identify the action of the excavator bucket and the surrounding environment features from the images of a large amount of excavator operation video data, so as to determine the loading stage to which the current time belongs. For example, when the trained ResNet34 model outputs the first shovel category, it indicates that the initial shovel loading action of the excavator is detected; when the trained ResNet34 model outputs the last shovel category, it indicates that the current shovel loading action is likely to be the last loading of the truck; when the trained ResNet34 model outputs the loading category, it indicates that the identification is an intermediate loading process between the first shovel and the last shovel; and when the trained ResNet34 model outputs the preparation state category, it indicates that the excavator is in the state of shovel, standby, or non-loading operation (for example, the transport truck has not been positioned or is in an intermittent state after completing a loading cycle). The surrounding environment features refer to: the weather conditions, various objects (other working excavators, queued trucks, moving trucks, retaining walls, and operation-related materials, etc.) and special environment (night low light) features in the surrounding environment.

[0041] In the embodiment, the state machine logic module 104 is further configured to set a plurality of state nodes of the excavator loading process, and the plurality of state nodes include an idle state, a start verification state, a loading state, and a cooling state.

[0042] In the embodiment, the state machine logic module 104 is specifically configured to determine that the state machine is in an idle state when the loading state classification result is a preparation state category; when a continuous first shovel category is detected in the idle state, the state machine is switched from the idle state to a start verification state; when a continuous loading-in-category is detected in the start verification state, the state machine is switched from the start verification state to a loading-in state, and the loading-in number is accumulated and the start time of loading of each vehicle is recorded; when a continuous last shovel category is detected in the loading-in state, the state machine is switched from the loading-in state to a cooling state, and the end time of loading of each vehicle is recorded, the loading duration of each vehicle is determined based on the start time of loading of each vehicle and the end time of loading of each vehicle, and a full load event of the vehicle is identified, and the state machine returns to the idle state and waits for the next loading period until all loading periods are completed.

[0043] In the embodiment, the vision-based shovel loading efficiency analysis and truck full load identification system further comprises an uploading module.

[0044] The uploading module is connected with the data recording and analysis module, and is configured to transmit the loading efficiency analysis data to the dispatch center in real time.

[0045] In the embodiment, the vision-based shovel loading efficiency analysis and truck full load identification system further comprises:

[0046] The display terminal interface is connected with the data recording and analysis module, and is configured to receive an audible and visual or state signal when the full load event of the vehicle is generated, and send the audible and visual or state signal to the cab display to remind that the current vehicle is full.

[0047] Figure 3 A working flowchart of the state machine logic module of the shovel is schematically shown, referring to Figure 3 As shown, specifically, the state machine logic module is configured to model and track the working state of the shovel according to the classification results continuously output by the AI classification module. In the initial condition, the state machine is in an idle state (corresponding to the preparation state category). Figure 3The state machine is switched from the idle state to the "start verification" state when the camera-captured image continuously shows the "first shovel" event. The state machine is switched from the "start verification" state to the "cooling" state when no continuous loading is identified within 1 minute; the state machine is switched from the "start verification" state to the "loading in progress" state when the camera-captured image continuously shows the "loading" event, and the following actions are performed: starting a new loading cycle (loading number count is increased by 1), and starting a timer to record the start time of loading of each vehicle. Subsequently, during the loading process, the state machine remains in the loading in progress state when the AI classification module outputs the "loading in progress" category, and the shovel loading number or the progress can be continuously accumulated internally, but the loading number is not repeated (the loading number is increased by 1 only at the first shovel). When the AI classification module identifies that the loading time of the current vehicle exceeds the maximum loading time, and the last shovel or the new vehicle signal is not identified, the state machine needs to be switched to the idle state for further execution; when the AI classification module continuously identifies the "last shovel", it means that the current transport vehicle is about to be fully loaded, and the state machine is switched from the loading in progress state to the "cooling" state. The state machine triggers the following operations: stopping the timer to obtain the total duration of the current loading operation, i.e., the loading time of each vehicle, and storing the total duration of the current loading operation and the corresponding loading number ID in the data recording and analysis module; and recording a full loading event.

[0048] When the full loading event is received, the application can generate an audible and visual signal or a status signal, and send the audible and visual signal or the status signal to the cab display to remind the driver that the current vehicle is fully loaded. After the state machine completes the above operation of the current vehicle being fully loaded, the state machine is reset to the idle state, and waits for the next transport vehicle to start a new loading cycle. Through the above state machine logic, the application realizes automatic measurement of the loading number and the loading bucket number, and accurate capture of the full loading time. The three states in the loading in progress state are full loading completion, non-full loading completion, and abnormal departure.

[0049] In the embodiment, the data recording and analysis module 105 is further configured to save the shovel number, the vehicle identifier, the loading sequence number, the start time of loading of each vehicle, and the end time of loading of each vehicle.

[0050] Specifically, the data recording and analysis module saves the data entries of each complete loading cycle, including the loading sequence number, start time, end time, and duration, etc. Optionally, the data recording and analysis module also counts the number of shovels used for each loading (i.e., the total number of shovels from the first shovel to the last shovel) and the corresponding vehicle identification (in the case of identifying the license plate number through vision or other means). The above-mentioned data saved by the data recording and analysis module can be transmitted to the dispatch center or cloud platform in real time, or the above-mentioned saved data can be finally exported to generate an efficiency analysis report. For example, from the provided efficiency analysis report, it can be obtained that how many times a certain excavator completes the loading operation within a certain shift, how long the average loading time per vehicle is, whether there is a significantly longer loading time (indicating possible abnormal stop or low efficiency), and the full load rate, etc. Based on the above data analysis, the working efficiency of the excavator and the driver can be better evaluated, and the vehicle scheduling and loading coordination can be optimized to improve the overall production efficiency.

[0051] The present application uses a pure visual method to obtain loading state information, has the characteristics of convenient installation and strong adaptability. The camera can be installed at the best viewing angle of the excavator as needed, and is equipped with dustproof, anti-vibration and night vision devices to adapt to the open environment. The edge computing device selects industrial-grade hardware and can operate stably in high temperature, severe cold, dust vibration and other environments. Since all intelligent analysis is completed locally, the dependence of the vision-based excavator loading efficiency analysis and truck full load recognition system on the communication network is minimized, so it can also play a role in poor network construction environment. In addition, the software module of the vision-based excavator loading efficiency analysis and truck full load recognition system of the present application has scalability. For example, new recognition categories (such as detecting the "underload" state, or recognizing different material types, etc.) can be added by updating the model of the AI classification module, and the logic of the state machine logic module can be adjusted to adapt to different operation processes. As can be seen, the vision-based excavator loading efficiency analysis and truck full load recognition system provided by the present application can be flexibly applied to various excavator loading scenes, providing an intelligent and efficient auxiliary tool for engineering construction and mining operations; it can automatically complete the key event recognition and data statistical analysis of the loading process, and has a significant advantage in improving operation safety and efficiency.

[0052] Compared with the prior art, the present application has the following advantages:

[0053] 1. Pure visual loading monitoring: The present application is based on camera image acquisition and recognition, realizing full-process monitoring of the loading state of the excavator, without the need for additional hardware such as weight sensors, position sensors, etc. to complete loading counting and full load recognition, reducing system complexity and cost.

[0054] 2. Specific four-classification model: The application adopts a four-classification deep learning model trained for the loading condition of the excavator, i.e., an image recognition model of a deep convolutional neural network (such as ResNet34), which can accurately distinguish the key stages (first shovel, loading, last shovel) in the loading process and the non-working state (preparation state), is more accurate in recognition than the existing general model, reduces misjudgment and omission, and improves the detection reliability.

[0055] 3. State machine logic framework: By setting a special state machine logic, the application converts the continuous loading state classification result into a stage determination of the loading process, and can trigger corresponding events (such as counting or timing) based on the recognized state. The state machine logic framework effectively ensures the coherence of the understanding and processing of the loading process, avoids the influence of transient errors in single-frame recognition on the overall judgment, and thus realizes stable automatic metering and event judgment.

[0056] 4. High robustness and real-time edge deployment: The application is run in real time on the edge computing device of the excavator, uses AI acceleration hardware such as GPU / NPU and optimized software model to realize low-delay inference and decision-making. Even in harsh outdoor environments, the visual-based excavator loading efficiency analysis and truck full load recognition system of the application still has strong robustness and real-time performance, and can adapt to the needs of operation monitoring under different light and dust conditions. At the same time, the local processing of the application avoids the delay of wireless network transmission, has low dependence on the network, and ensures that data is obtained in time and reliably.

[0057] From the above analysis, it can be seen that the visual-based excavator loading efficiency analysis and truck full load recognition system provided by the application can automatically and efficiently supervise the loading operation process of the excavator, realize loading frequency statistics, loading bucket number statistics, full load automatic recognition and loading time recording, and the generated analysis data can be used for optimizing operation scheduling and improving construction efficiency, and has a broad application prospect.

[0058] Although some specific embodiments of the application have been described in detail through examples, those skilled in the art should understand that the above examples are only for illustration, not for limiting the scope of the application. Those skilled in the art should understand that the above embodiments can be modified or some technical features can be replaced equivalently without departing from the scope and spirit of the application. In particular, as long as there is no structural conflict, each technical feature mentioned in each embodiment can be combined in any way.

[0059] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A vision-based analysis system for analyzing loading efficiency of a shovel and identifying a truck as fully loaded, the system comprising: The system comprises: a camera mounted on the excavator for collecting a real-time video stream of a working site of the excavator; an edge computing device connected with the camera for receiving the real-time video stream; an AI classification module deployed on the edge computing device for inputting the real-time video stream into a trained deep convolutional neural network image recognition model to classify the real-time video stream by using the trained deep convolutional neural network image recognition model, and outputting a loading state classification result, the loading state classification result comprising a first shovel category, a loading-in-progress category, a last shovel category, and a preparation state category; a state machine logic module connected with the AI classification module for state discrimination and state conversion according to the loading state classification result, and accumulating the number of loading times, identifying a full loading event of a truck, and recording the loading duration of each truck; a data recording and analysis module connected with the state machine logic module for saving the number of loading times, the full loading event of the truck, and the loading duration of each truck to generate loading efficiency analysis data, the loading efficiency analysis data being used for analyzing the loading efficiency of the excavator, optimizing vehicle scheduling, and loading coordination; the state machine logic module is further configured to set a plurality of state nodes of an excavator loading process, the plurality of state nodes comprising an idle state, a start verification state, a loading-in-progress state, and a cooling state; the state machine logic module is specifically configured to determine that the state machine is in the idle state when the loading state classification result is the preparation state category; when a continuous first shovel category is detected in the idle state, the state machine is converted from the idle state to the start verification state; when a continuous loading-in-progress category is detected in the start verification state, the state machine is converted from the start verification state to the loading-in-progress state, and the number of loading times is accumulated and the start time of loading of each truck is recorded; when a continuous last shovel category is detected in the loading-in-progress state, the state machine is converted from the loading-in-progress state to the cooling state, and the end time of loading of each truck is recorded, the loading duration of each truck is determined based on the start time of loading of each truck and the end time of loading of each truck, and the full loading event of the truck is identified, the state machine returns to the idle state and waits for the next loading period until all loading periods are completed.

2. The vision-based analysis system for shovel loading efficiency and truck fullness identification of claim 1, wherein, The trained deep convolutional neural network image recognition model comprises a trained ResNet34 model, the trained ResNet34 model being a model obtained by training a ResNet34 model by using a plurality of excavator working video data and a label with a state of each frame, the label with the state of each frame comprising a label of the first shovel category, a label of the loading-in-progress category, a label of the last shovel category, and a label of the preparation state category.

3. The vision-based shovel loading efficiency analysis and truck fullness identification system according to claim 1, characterized in that, The excavator loading efficiency analysis and truck full loading identification system based on vision further comprises an uploading module; the uploading module is connected with the data recording and analysis module for real-time transmission of the loading efficiency analysis data to a dispatch center.

4. The vision-based shovel loading efficiency analysis and truck fullness identification system of claim 1, wherein, The visual-based excavator loading efficiency analysis and truck full-load identification system further comprises: The display terminal interface is connected with the data recording and analysis module, and is used for receiving an audible and visual or state signal generated when the truck full-load event occurs, and sending the audible and visual or state signal to a cab display to remind that the current vehicle has been fully loaded.

5. The vision-based shovel loading efficiency analysis and truck fullness identification system according to claim 1, wherein, The data recording and analysis module is further used for saving the number of times of the shovel, the vehicle identification, the truck loading sequence number, the start time of loading of each vehicle, and the end time of loading of each vehicle.

6. The vision-based analysis system for shovel loading efficiency and truck fullness identification of claim 1, wherein, The edge computing device is an industrial computer, an embedded AI acceleration terminal, or a software and hardware integrated device.

7. The vision-based shovel loading efficiency analysis and truck fullness identification system of claim 1, wherein, An artificial intelligence hardware acceleration unit is arranged in the edge computing device, and the artificial intelligence hardware acceleration unit is a graphics processor or a neural network acceleration chip.

Images