Industrial gas cylinder placement state detection method and system based on machine vision

Abstract

The embodiment of the invention relates to the technical field of deep learning and target detection, and discloses an industrial gas cylinder placement state detection method and system based on machine vision. The advanced deep learning technology is adopted, the industrial gas cylinder and the empty cart can be efficiently and accurately recognized through the cascading mode of the industrial gas cylinder detection model and the cylinder head detection model, the detection precision and efficiency are improved, the secondary detection of the cylinder head is carried out for the target image of the empty cart, and the detection accuracy and efficiency are improved. Whether bottles exist on the cart or not is further ensured, and the accuracy of the system is enhanced. And when the detection result is null or a small target exists, the image is automatically segmented and amplified, so that the small target can be effectively identified, and the detection robustness is improved. And through post-processing steps such as edge filtering, overlapping frame filtering and category filtering, the detection result is further optimized, false alarm and missing alarm are reduced, and the reliability of the system is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep learning and target detection, and in particular, relates to a method and system for detecting the placement status of industrial gas cylinders based on machine vision. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No description herein is admitted to be prior art by inclusion in this section.

[0003] With the continuous development of industrial production, the frequency of use of gas cylinders has gradually increased, especially the widespread use of oxygen and acetylene cylinders in processes such as welding and cutting. However, if the gas cylinder is placed upside down during use, it may cause safety hazards such as gas leakage and bottle damage. Therefore, it is particularly important to detect the placement status of the gas cylinder in a timely and accurate manner.

[0004] Traditional gas cylinder inspection methods usually rely on manual inspections, which is not only inefficient but also easily affected by human factors, resulting in missed inspections or misjudgments. Therefore, the existing technology needs to be developed. Summary of the invention

[0005] Based on the above problems in the prior art, the present invention realizes efficient and accurate identification of the placement status of gas cylinders in industrial sites through a method and system for detecting the placement status of industrial gas cylinders based on machine vision.

[0006] In order to achieve the above object, the present invention adopts the following technical solution.

[0007] First, on the one hand, the present invention provides a method for detecting the placement status of industrial gas cylinders based on machine vision, the method comprising: inputting an image to be detected into an industrial gas cylinder detection model, detecting industrial gas cylinders and / or empty vehicles in the image, and obtaining gas cylinder / empty vehicle detection results, the gas cylinder / empty vehicle detection results comprising an upright category, an inverted category, an empty vehicle category, and a bottle on vehicle category; and outputting the detection results.

[0008] In some embodiments, the method further includes: performing bottle head detection on an image whose gas cylinder / empty vehicle detection result is an empty vehicle category, and if the bottle head detection result is that there is a bottle head, outputting the bottle is on the vehicle category.

[0009] In some embodiments, the step of inputting the image to be detected into an industrial gas cylinder detection model, detecting industrial gas cylinders and / or empty trucks in the image, and obtaining a gas cylinder / empty truck detection result includes: inputting the image to be detected into an industrial gas cylinder detection model, detecting industrial gas cylinders and / or empty trucks in the image, and obtaining a first detection result; when the first detection result is empty or there are small targets, dividing the image into several small images for secondary detection to obtain a secondary detection result; and comparing the first detection result with the secondary detection result to obtain a gas cylinder / empty truck detection result.

[0010] In some implementations, outputting the detection result includes: outputting after edge filtering, overlapping frame filtering and / or category filtering.

[0011] In some embodiments, the method also includes: constructing an industrial gas cylinder dataset D1; constructing a bottle head detection dataset D2; dividing the industrial gas cylinder dataset D1 and the bottle head detection dataset D2 into a training set and a validation set according to a preset ratio, respectively, training the neural network, and obtaining an industrial gas cylinder detection model and a bottle head detection model.

[0012] In some embodiments, constructing the industrial gas cylinder dataset D1 includes: collecting industrial gas cylinder data at the industrial site, cleaning the dataset, annotating the images, and constructing the industrial gas cylinder dataset D1, wherein the annotated categories include four categories: upright category, inverted category, bottle on vehicle category, and empty vehicle category. The upright category includes industrial gas cylinders in an upright state, industrial gas cylinders leaning against a wall, or on an object at an angle greater than or equal to 60° to the ground; the inverted category includes industrial gas cylinders inverted on the ground, industrial gas cylinders leaning against a wall, or on an object at an angle less than 60° to the ground; the empty vehicle category includes an empty vehicle without a cover, an empty vehicle with a cover, and a trailer with a cover but blocking the bottle.

[0013] In some embodiments, constructing the bottle head detection dataset D2 includes: cutting out a target image from an image of an empty vehicle category in the dataset D1, and labeling the bottle head category, to construct the bottle head dataset D2, wherein the target image includes a trailer and an area on the trailer for placing bottles.

[0014] On the other hand, the present invention also provides an industrial gas cylinder placement status detection system based on machine vision, which is used to implement the above-mentioned industrial gas cylinder placement status detection method. The system includes: a graphic acquisition module, which is used to acquire the image to be detected; an industrial gas cylinder detection module, which is provided with an industrial gas cylinder detection model, which is used to detect industrial gas cylinders and / or empty vehicles in the image to obtain gas cylinder / empty vehicle detection results, and the gas cylinder / empty vehicle detection results include upright categories, inverted categories, empty vehicle categories and bottle on vehicle categories; an output module, which is used to output the detection results.

[0015] In some embodiments, the system further includes a bottle head detection module, which is provided with a bottle head detection model for performing bottle head detection on images whose gas cylinder / empty vehicle detection result is an empty vehicle category. If the bottle head detection result is that there is a bottle head, the bottle is on the vehicle category is output.

[0016] In some implementations, outputting the bottle-on-vehicle category includes modifying the gas cylinder / empty vehicle detection result of the image into the bottle-on-vehicle category, and outputting the result.

[0017] Beneficial effects of this application:

[0018] (1) This application adopts advanced deep learning technology. By cascading the industrial gas cylinder detection model and the bottle head detection model, on the one hand, it can efficiently and accurately identify industrial gas cylinders and empty carts, thereby improving the accuracy and efficiency of detection. On the other hand, it performs secondary detection of the bottle head on the target image of the empty cart to further ensure whether there are bottles on the cart, thereby enhancing the accuracy of the system. When the detection result is empty or there is a small target, the image is automatically segmented and enlarged to ensure that the small target can also be effectively identified, thereby improving the robustness of the detection. Through post-processing steps such as edge filtering, overlapping frame filtering, and category filtering, the detection results are further optimized, false positives and false negatives are reduced, and the reliability of the system is improved. In short, this application can automatically identify and judge the placement status of gas cylinders in real time, significantly improve detection efficiency and accuracy, reduce safety risks, meet the high requirements of modern industrial production for safety management, and at the same time. By combining advanced deep learning technology, the system can operate stably under different environmental conditions and adapt to various industrial application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0020] Figure 1 This is an overall flow chart of a method for detecting the placement status of industrial gas cylinders based on machine vision according to Embodiment 1 of the present invention;

[0021] Figure 2 This is a schematic diagram of the overall process of the prediction stage of Example 1 of the present invention;

[0022] Figure 3 This is a graph showing the output results of the industrial gas cylinder detection model according to Embodiment 1 of the present invention;

[0023] Figure 4 This is a graph of the output results of the bottle head detection model in Example 1 of the present invention. DETAILED DESCRIPTION

[0024] With the rapid development of deep learning and computer vision technology, intelligent detection systems based on image processing have gradually become an effective means to solve the problem of safety hazards caused by gas cylinders being placed upside down during use.

[0025] The YOLO series of algorithms is one of the mainstream algorithms in the current target detection field. As its latest version, YOLOv11 further improves detection accuracy and speed. YOLOv11 uses advanced network structure and optimization technology to enable real-time target detection in complex scenarios and effectively handle the recognition of multiple categories of objects. The algorithm significantly improves the detection speed by converting target detection into a regression problem, which is suitable for rapid judgment of the placement status of gas cylinders in industrial environments.

[0026] In this context, the present invention proposes a method and system for detecting the placement status of industrial gas cylinders based on machine vision. The method uses deep learning technology to automatically identify and judge the placement status of gas cylinders in real time, significantly improve detection efficiency and accuracy, reduce safety risks, and meet the high requirements of modern industrial production for safety management. By combining advanced deep learning technology, the system can operate stably under different environmental conditions and adapt to various industrial application scenarios.

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0028] It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments, which also fall within the scope of protection of the present invention.

[0029] In the description of the present invention, the following terms need to be explained:

[0030] As for directional words, the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicating directions and positional relationships in the specification and claims of this application are based on the directions or positional relationships shown in the drawings and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and should not be understood as limiting the specific scope of protection of the present invention.

[0031] The terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects and are not necessarily used to describe a particular sequence or order.

[0032] The terms "including" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0033] When an element is referred to as being "fixed to" or "disposed on" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. When an element is referred to as being "provided with" another element, it may be provided on the surface or inside of the element.

[0034] Unless explicitly stated otherwise, throughout the specification and claims, the term “comprise” or variations such as “include” or “comprising”, etc., will be understood to include the stated elements or components but not to exclude other elements or components.

[0035] The term "electrical connection" in the specification and claims of this application may refer to a circuit connection with physical contact or a communication connection, which may be a wired communication connection or a wireless communication connection.

[0036] Example 1

[0037] First, this embodiment provides a method for detecting the placement status of industrial gas cylinders based on machine vision, the method comprising: inputting an image to be detected into an industrial gas cylinder detection model, detecting industrial gas cylinders and / or empty trucks in the image, and obtaining gas cylinder / empty truck detection results, the gas cylinder / empty truck detection results including an upright category, an inverted category, an empty truck category, and a bottle-on-truck category; and outputting the detection results.

[0038] Specifically, the above scheme is implemented by the following steps, and the overall flow chart of the method is as follows: Figure 1 As shown:

[0039] Step 1: Preprocess the image, annotate the image, construct the industrial gas cylinder dataset D1, and construct the bottle head detection dataset D2;

[0040] Specifically, the industrial gas cylinder dataset D1 is first constructed in the following way: industrial gas cylinder data are collected from industrial sites, the dataset is cleaned, and the images are annotated using the labelImg annotation tool. The annotation categories include upright, inverted, bottles on the vehicle, and empty vehicles. The upright category includes industrial gas cylinders in an upright state, industrial gas cylinders leaning against a wall, or on objects with an angle greater than or equal to 60° to the ground; the inverted category includes industrial gas cylinders inverted on the ground, industrial gas cylinders leaning against a wall, or on objects with an angle less than 60° to the ground; the empty vehicle category includes empty vehicles without covers, empty vehicles with covers, and trailers with covers but blocking the bottles. After the industrial gas cylinder dataset D1 is constructed, the target image of the empty vehicle category in the dataset D1 image is cut out, and the bottle head category is marked to construct the bottle head dataset D2. The target image includes the trailer and the area on the trailer for placing bottles. That is, the cut-out target image should include the trailer. If it is a bottle upstream of the trailer, it should also include the bottle. If the trailer is empty, it should also include the area on the trailer for placing bottles or the area that should be covered if the trailer is placed with bottles. In short, the cut-out target image should be able to distinguish whether the trailer is empty or has bottles on it. According to actual conditions, the method for constructing the bottle head dataset D2 can also not construct the bottle head dataset D2 based on D1, but re-collect the industrial gas cylinder bottle head data at the industrial site, as well as the bottle heads of the bottles taken in various placement states, angles, and directions on the vehicle, and the images of the bottle heads in different states such as open, closed, and loose, so as to increase the richness of the dataset.

[0041] Step 2: Use a deep learning neural network to train an industrial gas cylinder detection model. In this embodiment, YOLOV11 is used to train a bottle head detection model.

[0042] Step 2.1: First, divide the industrial gas cylinder dataset D1 into a training set and a validation set in a ratio of 8:2;

[0043] Step 2.2: Input the industrial gas cylinder data set D1 into the YOLOV11 network for training, train the model, and obtain a trained neural network, wherein the YOLOV11 network uses the SIOU loss function to calculate the positioning loss and the classification loss;

[0044] Step 2.3: When the neural network converges, the trained neural network is determined as an industrial gas cylinder detection model.

[0045] Step 2.4: Divide the bottle head data set D2 into a training set and a validation set in a ratio of 8:2.

[0046] Step 2.5: Input the bottle head data set D2 into the YOLOV11 network for training, train the model, and obtain the trained neural network.

[0047] Step 2.6: When the neural network converges, the trained neural network is determined as a bottle head detection model.

[0048] It should be noted that the training order of the above industrial gas cylinder detection model and the bottle head detection model does not have to follow the above order and can be flexibly adjusted according to actual conditions.

[0049] Step 3: In the prediction stage, the industrial gas cylinder detection model is first called to detect all industrial gas cylinders and empty carts in the image to obtain the first detection result. In order to prevent missed detection or false detection when the target is too small, when the first detection result is empty or there is a small target, this embodiment divides the image into 4 small images of the top, bottom, left and right and enlarges them, performs a secondary detection, and obtains the secondary detection result; the first detection result and the secondary detection result are compared to obtain the gas cylinder / empty cart detection result. The overall flow chart of the prediction stage is as follows: Figure 2 shown.

[0050] The specific method of step 3 is:

[0051] Step 3.1: Set the industrial gas cylinder detection parameters, which include small target detection parameter a and bottle head detection parameter b, where a, b∈{0,1};

[0052] Step 3.3: Input the image into the industrial gas cylinder detection model and output the first detection result. The first detection result is result = [result1, result2, ..., resulti, ..., resultn], where n is the number of all target detection boxes in the image, resulti = [x1, y1, x2, y2, conf1, class_label1], x1, y1, x2, y2 are the upper left corner x coordinate of the detection box, the upper left corner y coordinate of the detection box, the lower right corner x coordinate of the detection box, and the lower right corner y coordinate of the detection box, respectively, and conf1 is the confidence and the label corresponding to the category name of the class_label1 object. class_label1∈{0,1,2,3}, 0 represents bottle_norm (upright category), 1 represents bottle_down (inverted category), 2 represents cart (empty car category), and 3 represents bottle_on_cart (bottle on car category); the output results of the industrial gas cylinder detection model are as follows: Figure 3 shown.

[0053] Step 3.4: If a is equal to 1, jump to step 3.5, otherwise jump to step 3.5.11;

[0054] Step 3.5: If result is empty, jump to step 3.5.1, otherwise jump to step 3.5.11;

[0055] Step 3.5.1: w = x2-x1, h = y2-y1;

[0056] Step 3.5.2: If w < 30 pixels or h < 30 pixels, go to step 3.5.3, otherwise jump to step 3.5.11;

[0057] Step 3.5.3: Cut the original image into 4 sub-images (upper left, upper right, lower left, and lower right), enlarge the sub-images to the size of the original image, and use linear interpolation to enhance the features of the sub-images;

[0058] Step 3.5.4: Input the industrial gas cylinder detection model again and output the detection result small_result, where small_result = [small_result1, small_result2, ..., small_resulti, ..., small_resultn], where n is the number of all detection frames in the image, small_resulti = [x01, y01, x02, y02, conf2, class_label2], x01, y01, x02, y02 are the upper left corner x coordinate of the detection frame, the upper left corner y coordinate of the detection frame, the lower right corner x coordinate of the detection frame, the lower right corner y coordinate of the detection frame, conf2 is the confidence, and class_label2 is the label corresponding to the class name of the object. class_label2∈{0,1,2,3}, 0 represents bottle_norm (upright category), 1 represents bottle_down (inverted category), 2 represents cart (empty car category), and 3 represents bottle_on_cart (bottle on car category);

[0059] Step 3.5.5: If small_result is empty, go to step 3.5.11, otherwise go to step 3.5.6;

[0060] Step 3.5.6: Convert the detected target frame coordinates from the segmented sub-image area coordinate system back to the original image coordinate system as the secondary detection result small_result;

[0061] Step 3.5.7: Compare the secondary detection result small_result with the first detection result result, and calculate the intersection-over-union (IOU) ratio between all target boxes;

[0062] Step 3.5.8: If IOU>0.6, jump to step 3.5.9, otherwise jump to step 3.5.10;

[0063] Step 3.5.9: Keep the results with high confidence;

[0064] Step 3.5.10: Keep the test results of both tests;

[0065] Step 3.5.11: Output the result bottle_result as the cylinder / empty vehicle detection result.

[0066] Step 4: Perform secondary detection on the empty cart target image to detect the bottle head and further determine whether there is a bottle on the cart. Specifically, step 4 includes the following steps:

[0067] Step 4.1: If b=1, jump to step 4.2, otherwise jump to step 5.1;

[0068] Step 4.2: If class_label = 2, jump to step 4.3, otherwise jump to step 5.1;

[0069] Step 4.3: Cut the target image out of the original image and input it into the bottle head detection model, output the detection result head_result, head_result = [head_result1, head_result2, ... head_resulti, ..., head_resultn], where n is the number of all detection boxes in the image, small_resulti = [x001, y001, x002, y002, conf3, class_label3], x001, y001, x002, y002 are the upper left corner x coordinate of the detection box, the upper left corner y coordinate of the detection box, the lower right corner x coordinate of the detection box, the lower right corner y coordinate of the detection box, and class_label3 is the label corresponding to the category name of the object. class_label3∈{0}, 0 means head (bottle head class).

[0070] Step 4.4: If head_result is not empty, go to step 4.5, otherwise go to step 5.1;

[0071] Step 4.5: Change the detection result category class_label to 3. The output of the bottle head detection model is as follows Figure 4 shown.

[0072] Step 5: Process the final reported result, perform edge filtering, overlapping frame filtering, and category filtering, and output the final result. Specifically, step 5 may include the following steps:

[0073] Step 5.1: Perform bounding box filtering on the results;

[0074] Step 5.1.1: Set the edge tolerance edge_margin and the minimum aspect ratio threshold mini_ratio, where edge_margin represents the pixel value from the left and right boundaries of the image, and the minimum aspect ratio threshold mini_ratio is set;

[0075] Step 5.1.2: w = x20-x10, h = y20-y10, where x10 and x20 are the x coordinates of the upper left corner and the lower right corner of the target box, and y10 and y20 are the y coordinates of the upper left corner and the lower right corner;

[0076] Step 5.1.3: If x10 is less than or equal to the edge tolerance edge_margin or x20 is greater than or equal to the image width minus the edge tolerance, the target box is considered to be a bounding box;

[0077] Step 5.1.4: Calculate the aspect ratio of the bounding box aspect_ratio = h / w. If aspect_ratio is less than the set minimum aspect ratio threshold mini_ratio, delete the target box, otherwise jump to step 5.3.5;

[0078] Step 5.1.5: Return result;

[0079] Step 5.2: Filter the results by overlapping frames;

[0080] Step 5.2.1: Find the IOU between all target boxes in the result and other target boxes; if iou>0.8, jump to step 5.2.2;

[0081] Step 5.2.2: If the target box categories are the same, delete the target box with a smaller confidence conf, otherwise jump to step 5.2.3;

[0082] Step 5.2.3: If the target box category is 2 or 3, delete the target box corresponding to 3. If it is other categories, delete the target box with small confidence.

[0083] Step 5.3: Filter the results by category;

[0084] Step 5.3.1: Process the input category. If the category is 3, change the category to 1.

[0085] Step 5.3.2: Map the category to the original label name, perform category filtering, report the bottle_norm and bottle_down labels, and do not report the cart category.

[0086] Step 5.4: Output the final result.

[0087] On the other hand, this embodiment also provides an industrial gas cylinder placement status detection system based on machine vision, which is used to implement the above-mentioned industrial gas cylinder placement status detection method. The system includes: a graphics acquisition module, which is used to acquire the image to be detected; an industrial gas cylinder detection module, which is provided with an industrial gas cylinder detection model, which is used to detect industrial gas cylinders and / or empty vehicles in the image to obtain gas cylinder / empty vehicle detection results, and the gas cylinder / empty vehicle detection results include upright categories, inverted categories, empty vehicle categories and bottle on vehicle categories; an output module, which is used to output the detection results.

[0088] In some embodiments, the system further includes a bottle head detection module, which is provided with a bottle head detection model for performing bottle head detection on images whose gas cylinder / empty vehicle detection result is an empty vehicle category. If the bottle head detection result is that there is a bottle head, the bottle is on the vehicle category is output.

[0089] In some implementations, outputting the bottle-on-vehicle category includes modifying the gas cylinder / empty vehicle detection result of the image into the bottle-on-vehicle category, and outputting the result.

[0090] The method and system for detecting the placement status of industrial gas cylinders based on machine vision proposed in the present invention have the following beneficial effects:

[0091] 1. The detection model trained by YOLOv11 can efficiently and accurately identify industrial gas cylinders and empty carts, improving the accuracy and efficiency of detection.

[0092] 2. When the detection result is empty or there is a small target, the image is automatically segmented and enlarged to ensure that the small target can also be effectively identified, thereby improving the robustness of the detection.

[0093] 3. Perform secondary inspection of bottle heads on the target image of an empty cart to further ensure whether there are bottles on the cart, enhancing the accuracy of the system.

[0094] 4. Through post-processing steps such as edge filtering, overlapping box filtering and category filtering, the detection results are further optimized, false positives and false negatives are reduced, and the reliability of the system is improved.

[0095] 5. The modular design of this method allows it to be easily integrated into other machine vision systems, and has a wide range of application potential and adaptability. At the same time, by real-time monitoring of the placement status of gas cylinders, potential risks can be discovered in a timely manner, the safety of the production environment can be enhanced, and the probability of accidents can be reduced.

[0096] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents, and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the placement status of industrial gas cylinders based on machine vision, characterized in that: The method comprises: inputting an image to be detected into an industrial gas cylinder detection model, detecting industrial gas cylinders and / or empty vehicles in the image, obtaining gas cylinder / empty vehicle detection results, wherein the gas cylinder / empty vehicle detection results include an upright category, an inverted category, an empty vehicle category, and a bottle-on-vehicle category; and outputting the detection results.

2. The method for detecting the placement status of industrial gas cylinders based on machine vision according to claim 1 is characterized in that: The method further includes: performing bottle head detection on an image whose gas cylinder / empty vehicle detection result is an empty vehicle category, and if the bottle head detection result is that there is a bottle head, outputting the bottle is on the vehicle category.

3. The method for detecting the placement status of industrial gas cylinders based on machine vision according to claim 2 is characterized in that: The step of inputting the image to be detected into the industrial gas cylinder detection model, detecting the industrial gas cylinder and / or empty vehicle in the image, and obtaining the gas cylinder / empty vehicle detection result comprises: inputting the image to be detected into the industrial gas cylinder detection model, detecting the industrial gas cylinder and / or empty vehicle in the image, obtaining a first detection result, and when the first detection result is empty or there is a small target, dividing the image into a plurality of small images for secondary detection, and obtaining a secondary detection result; The first detection result and the second detection result are compared to obtain the gas cylinder / empty vehicle detection result.

4. The method for detecting the placement status of industrial gas cylinders based on machine vision according to claim 3 is characterized in that: The output detection result includes: outputting after edge filtering, overlapping frame filtering and / or category filtering.

5. The method for detecting the placement status of industrial gas cylinders based on machine vision according to claim 4 is characterized in that: The method further comprises: Construct industrial gas cylinder dataset D1; Construct bottle head detection dataset D2; The industrial gas cylinder data set D1 and the bottle head detection data set D2 are divided into a training set and a validation set according to a preset ratio, and the neural network is trained to obtain the industrial gas cylinder detection model and the bottle head detection model.

6. The method for detecting the placement status of industrial gas cylinders based on machine vision according to claim 5 is characterized in that: The construction of the industrial gas cylinder dataset D1 includes: Industrial gas cylinder data from industrial sites were collected, the data set was cleaned, the images were annotated, and the industrial gas cylinder data set D1 was constructed. The annotated categories include upright category, inverted category, bottle on vehicle category, and empty vehicle category. The upright category includes industrial gas cylinders in an upright state, industrial gas cylinders leaning against a wall or on an object with an angle greater than or equal to 60° to the ground; the inverted category includes industrial gas cylinders inverted on the ground, industrial gas cylinders leaning against a wall or on an object with an angle less than 60° to the ground; the empty vehicle category includes empty vehicles without covers, empty vehicles with covers, and trailers with covers but blocking the bottles.

7. The method for detecting the placement status of industrial gas cylinders based on machine vision according to claim 5 is characterized in that: The construction of the bottle head detection dataset D2 includes: cutting out the target image from the image of the empty vehicle category in the industrial gas cylinder dataset D1, marking the bottle head category, and constructing the bottle head dataset D2, wherein the target image includes the trailer and the area on the trailer for placing bottles.

8. A machine vision-based industrial gas cylinder placement status detection system, used to implement the industrial gas cylinder placement status detection method according to any one of claims 1 to 7, characterized in that: include: A graphics acquisition module, used for acquiring an image to be detected; An industrial gas cylinder detection module is provided with an industrial gas cylinder detection model, which is used to detect industrial gas cylinders and / or empty vehicles in an image and obtain gas cylinder / empty vehicle detection results, wherein the gas cylinder / empty vehicle detection results include upright categories, inverted categories, empty vehicle categories, and categories where the cylinder is on the vehicle; The output module is used to output the detection results.

9. The machine vision-based industrial gas cylinder placement status detection system according to claim 8 is characterized in that: It also includes a bottle head detection module, which is provided with a bottle head detection model for performing bottle head detection on images whose gas cylinder / empty vehicle detection result is an empty vehicle category. If the bottle head detection result is that there is a bottle head, the bottle is on the vehicle category is output.

10. The machine vision-based industrial gas cylinder placement status detection system according to claim 9, characterized in that: The outputting of the bottle-on-vehicle category includes modifying the gas cylinder / empty vehicle detection result of the image into the bottle-on-vehicle category and outputting it.

Images