Dangerous chemical identification system based on deep learning and automatic image identification and classification

Abstract

The invention provides a dangerous chemical identification system based on deep learning and automatic image identification and classification, which comprises a plurality of functional modules, obtains a plurality of view images of a target area in the identification process, determines a target image for classification and identification by carrying out operations such as interception and enhancement on the plurality of view images, and carries out classification and classification on the target image. And inputting the target image into a deep learning model, processing the target image through the deep learning model, extracting object features of each target object contained in the target image, identifying whether the target object is a dangerous chemical or not based on the object features of each target object, and when the target object is identified as the dangerous chemical, determining whether the target object is the dangerous chemical. And outputting the chemical types of the dangerous chemicals. By applying the system provided by the invention, when the dangerous chemicals are identified and classified, the image shot in the to-be-identified area is input into the deep learning model, whether the dangerous chemicals exist in the related area or not and related classification can be obtained through processing of the deep learning model, and the efficiency and accuracy of dangerous chemical identification are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a hazardous chemical identification system based on deep learning and automatic image recognition and classification. Background Art

[0002] Hazardous chemicals refer to highly toxic chemicals and other chemicals that are harmful to humans, facilities, and the environment due to their toxic, corrosive, explosive, flammable, and combustion-supporting properties. In view of the harmfulness of hazardous chemicals, they need to be identified promptly in public safety areas such as daily life and transportation.

[0003] Traditional methods of identifying hazardous chemicals mostly rely on manual inspection or handheld detection equipment, which have problems such as low efficiency and high misjudgment rate. Summary of the invention

[0004] In view of this, the present invention provides a hazardous chemical identification system based on deep learning and automatic image recognition and classification, which can quickly identify and classify hazardous chemicals, improve recognition efficiency, and reduce the misjudgment rate.

[0005] A hazardous chemical identification system based on deep learning and automatic image recognition and classification, comprising:

[0006] A receiving module, used to receive images of various viewing angles obtained by the detection device shooting the target area at multiple angles;

[0007] A selection module, used for selecting a viewing angle image taken at a normal viewing angle as a reference image from among the viewing angle images;

[0008] A determination module, used for determining an object area included in the reference image according to each of the view images;

[0009] A capture module, used to capture the image where the object area is located as the image to be recognized;

[0010] An enhancement module, used to perform enhancement processing on the image to be identified to obtain a target image corresponding to the image to be identified;

[0011] A recognition module, used to input the target image into an established deep learning model, determine the target objects contained in the target image and the object form of each target object through processing by the deep learning model, and extract the object features of each target object according to the object form of each target object, identify whether the target object is a hazardous chemical based on the object features of each target object, and output the chemical type of the hazardous chemical to which the target object belongs when the target object is identified as a hazardous chemical;

[0012] A training module is used to determine multiple test images, which include images of various types of hazardous chemicals; enhance the multiple test images to obtain multiple sample images; determine each object contained in each of the sample images, and add an attribute label to each object in each of the sample images; extract object features of each object in each of the sample images, wherein the object features include at least the shape, color, texture and edge of the object; determine an initial learning model, take each object feature of each object in each of the sample images as input, take the attribute label of each object in each of the sample images as output, and train the initial learning model until the convergence function of the initial learning model is less than a preset threshold, completing the training of the initial learning model to establish the deep learning model; each of the attribute labels is used to characterize whether the object corresponding to it is a hazardous chemical, and when the object is a hazardous chemical, the chemical type of the hazardous chemical to which the object belongs.

[0013] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification may optionally further include:

[0014] The detection device is used to photograph the target area from multiple angles to obtain a viewing angle image at each angle.

[0015] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification, optionally, the determination module includes:

[0016] An analysis submodule, configured to perform object analysis on each of the view angle images to determine an object region where an object exists at each view angle;

[0017] The mapping submodule is used to map the object area where the object exists in each viewing angle to the reference image, and determine the object area contained in the reference image.

[0018] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification, optionally, the interception module includes:

[0019] A judging submodule, used for judging whether the object region contained in the reference image is a continuous region;

[0020] The interception submodule is used to intercept the images corresponding to the discrete object regions contained in the reference image respectively if the object region contained in the reference image is not a continuous region, and use the intercepted images as the images to be identified.

[0021] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification, optionally, the enhancement module includes:

[0022] The enhancement submodule is used to perform enhancement processing on the image to be identified, wherein the enhancement processing at least includes any one or a combination of any several of grayscale, binarization, denoising and contrast enhancement.

[0023] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification may optionally further include:

[0024] The display module is used to display to the user the identification result of whether the target object output by the deep learning model is a hazardous chemical, and when the target object is a hazardous chemical, the chemical type of the hazardous chemical of the target object.

[0025] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification, optionally, the training module includes:

[0026] The cleaning submodule is used to clean the multiple test images to remove blurred images and repeated images in the multiple test images.

[0027] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification may optionally further include:

[0028] The model testing module is used to test the established deep learning model to determine the processing performance of the deep learning model.

[0029] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification may optionally further include:

[0030] The model optimization module is used to optimize the established deep learning model to adjust the model structure and hyperparameters of the deep learning model.

[0031] The above-mentioned dangerous chemical identification system based on deep learning and automatic image recognition and classification may optionally further include:

[0032] The model deployment interface is used to connect the deep learning model to the cloud server or edge device to complete the deployment of the deep learning model in the cloud server or edge device.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] The present invention provides a hazardous chemical identification system based on deep learning and automatic image recognition and classification, comprising: a receiving module, used to receive various viewing angle images obtained by a detection device shooting a target area from multiple angles; a selection module, used to select a viewing angle image shot from a positive viewing angle as a reference image among the various viewing angle images; a determination module, used to determine an object area contained in the reference image based on the various viewing angle images; a capture module, used to capture an image where the object area is located as an image to be identified; an enhancement module, used to perform enhancement processing on the image to be identified to obtain a target image corresponding to the image to be identified; and a recognition module, used to input the target image into an established deep learning model, determine various target objects contained in the target image and the object form of each target object through processing by the deep learning model, extract the object features of each target object according to the object form of each target object, identify whether the target object is a hazardous chemical based on the object features of each target object, and after identifying the target object When the target object is a hazardous chemical, the chemical type of the hazardous chemical to which the target object belongs is output; a training module is used to determine multiple test images, wherein the multiple test images contain images of various types of hazardous chemicals; the multiple test images are enhanced to obtain multiple sample images; each object contained in each of the sample images is determined, and an attribute label is added to each object in each of the sample images; the object features of each object in each of the sample images are extracted, and the object features at least include the shape, color, texture and edge of the object; an initial learning model is determined, and the object features of each object in each of the sample images are used as input, and the attribute label of each object in each of the sample images is used as output, and the initial learning model is trained until the convergence function of the initial learning model is less than a preset threshold, and the training of the initial learning model is completed to establish the deep learning model; each of the attribute labels is used to characterize whether the object corresponding to it is a hazardous chemical, and when the object is a hazardous chemical, the chemical type of the hazardous chemical to which the object belongs. When applying the system provided by the present invention to identify and classify hazardous chemicals, the images taken of the area to be identified are input into the deep learning model. After processing by the deep learning model, it can be obtained whether there are hazardous chemicals in the relevant area and the relevant classification, thereby improving the efficiency and accuracy of hazardous chemical identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0036] Figure 1 A schematic diagram of the structure of a hazardous chemical identification system based on deep learning and automatic image recognition and classification provided by an embodiment of the present invention;

[0037] Figure 2 Another structural schematic diagram of a hazardous chemical identification system based on deep learning and automatic image recognition and classification provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] 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.

[0039] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0040] The present invention can be used in many general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, distributed computing environments including any of the above devices or devices, etc.

[0041] The embodiment of the present invention provides a dangerous chemical identification system based on deep learning and automatic image recognition and classification. The dangerous chemical identification system can be applied to the fields of public safety, industrial production, and transportation hubs. Cloud servers corresponding to the above fields can be set up. The structural diagram of the dangerous chemical identification system based on deep learning and automatic image recognition and classification provided by the embodiment of the present invention is as follows: Figure 1 As shown, including:

[0042] The receiving module 101 is used to receive images of various viewing angles obtained by the detection device shooting the target area at multiple angles;

[0043] In the embodiment of the present invention, the receiving module 101 receives in real time the images of various viewing angles obtained by the detection device through multi-angle shooting of the target area.

[0044] The detection device may be various types of camera devices. In the embodiment of the present invention, images of different formats acquired by different types of camera devices may be processed.

[0045] In an embodiment of the present invention, the target area can be a fixed area, such as a security inspection machine area in the field of public transportation, or a necessary area that vehicles need to pass through in the transportation field. The detection equipment is set at each set position of the target area. When the user's luggage, goods transported by vehicles, or individuals entering and leaving important places pass through the target area, the detection equipment takes multi-angle photos of the objects or individuals passing through the target area to obtain perspective images at each angle.

[0046] The receiving module 101 receives each viewing angle image obtained by the detection device, and transmits each viewing angle image to the selecting module 102 .

[0047] A selection module 102 is used to select a perspective image taken at a normal perspective as a reference image from among the perspective images;

[0048] In an embodiment of the present invention, the selection module 102 performs perspective analysis on each perspective image received by the receiving module 101, and determines the normal perspective orientation according to the scene where the current target area is located. In actual operation, such as in the security inspection area of ​​a station or airport, the normal shooting angle of the detection equipment from top to bottom can be determined as the normal viewing angle. In the process of transporting large goods, the angle at eye level with the large goods can also be set as the formal angle. In an embodiment of the present invention, the normal viewing angle is determined according to the application scenario where the target area is located.

[0049] The selection module 102 selects a perspective image that matches the determined front viewing angle from among the perspective images as a reference image according to the determined front viewing angle.

[0050] A determination module 103, configured to determine an object region included in the reference image according to each of the view images;

[0051] In an embodiment of the present invention, during the actual shooting process of the detection device, since the objects in the luggage or cargo are irregularly placed and there are occlusions, in the process of determining the object area in the reference image, the object area of ​​the object contained in the reference image can be comprehensively determined based on the perspective images of different perspectives.

[0052] In actual application, the embodiments of the present invention can identify hazardous chemicals in a variety of scenarios. For example, in infrared security inspections, due to the characteristics of infrared rays, the object area where the object exists as a whole can be identified at multiple angles.

[0053] In the process of image capture, such as capturing images of various objects exposed in the actual environment, there may be occluded areas. In this case, it is necessary to comprehensively consider the view images from various angles to determine the object area in the reference image.

[0054] A capture module 104, used to capture the image where the object area is located as an image to be recognized;

[0055] In the embodiment of the present invention, the capture module 104 captures the object region portion where the object exists in the reference image, removes the portion of the reference image that does not contain the object, and uses the image of the portion where the object exists as the image to be recognized.

[0056] The enhancement module 105 is used to perform enhancement processing on the image to be identified to obtain a target image corresponding to the image to be identified;

[0057] In an embodiment of the present invention, the enhancement module 105 performs enhancement processing on the image to be identified, improves the processability of the image to be identified, and obtains a target image corresponding to the image to be identified, where the target image is an image input into the established deep learning model for identification processing.

[0058] The recognition module 106 is used to input the target image into the established deep learning model, determine the target objects contained in the target image and the object form of each target object through processing by the deep learning model, and extract the object features of each target object according to the object form of each target object, identify whether the target object is a hazardous chemical based on the object features of each target object, and output the chemical type of the hazardous chemical to which the target object belongs when the target object is identified as a hazardous chemical;

[0059] In the embodiment of the present invention, the recognition module 106 inputs the target image into the established deep learning model. In the deep learning model, each target object contained in the target image is further determined. Through the object characteristics of different objects, each target object in the target image can be distinguished. At the same time, the deep learning model extracts the object features of each target object. Based on the features of each target object, it can be identified whether the target object is a hazardous chemical. Furthermore, when some target objects are identified as hazardous chemicals, the type of hazardous chemical to which the target object belongs is further determined.

[0060] The training module 107 is used to determine multiple test images, which include images of various types of hazardous chemicals; enhance the multiple test images to obtain multiple sample images; determine each object contained in each of the sample images, and add an attribute label to each object in each of the sample images; extract object features of each object in each of the sample images, and the object features include at least the shape, color, texture and edge of the object; determine an initial learning model, take each object feature of each object in each of the sample images as input, take the attribute label of each object in each of the sample images as output, and train the initial learning model until the convergence function of the initial learning model is less than a preset threshold, completing the training of the initial learning model to establish the deep learning model; each of the attribute labels is used to characterize whether the object corresponding to it is a hazardous chemical, and when the object is a hazardous chemical, the chemical type of the hazardous chemical to which the object belongs.

[0061] In an embodiment of the present invention, in the process of obtaining a deep learning model through training, a plurality of test images are first determined, each of which may include an image of a type of hazardous chemical. The test image acquisition process may be performed by acquiring images of hazardous chemicals in different scenes through test equipment such as a video camera or a camera or an infrared acquisition device.

[0062] In an embodiment of the present invention, the multiple test images may include various hazardous chemicals of all categories. After the multiple test images are enhanced, the sample images used for deep learning model training are determined.

[0063] An attribute label is added to each object in each sample image. For example, if the object in the sample image does not belong to hazardous chemicals, the attribute label may mark the name and classification of the object and indicate that the object does not belong to hazardous chemicals.

[0064] If there are hazardous chemicals in the objects included in the sample image, the attribute label of the object may mark the object as a hazardous chemical and mark the classification and degree of hazard of the hazardous chemical.

[0065] In an embodiment of the present invention, the object features of each object in each sample image are used as input, and the attribute label of the object in each sample image is used as output, and the determined initial model is trained until the convergence value of the convergence function of the initial model is less than a preset threshold, and the training of the initial model is terminated to obtain a deep learning model.

[0066] The hazardous chemical identification system based on deep learning and automatic image recognition and classification provided in an embodiment of the present invention, when applying the system provided by the present invention to identify and classify hazardous chemicals, the images taken of the area to be identified are input into the deep learning model. After processing by the deep learning model, it can be obtained whether there are hazardous chemicals in the relevant area and the relevant classification, thereby improving the efficiency and accuracy of hazardous chemical identification.

[0067] refer to Figure 2 Another detailed structural diagram of the hazardous chemical identification system based on deep learning and automatic image recognition and classification provided by an embodiment of the present invention is shown. Figure 2 The system provided by the embodiment of the present invention further includes:

[0068] The detection device 108 is used to photograph the target area from multiple angles to obtain a viewing angle image at each angle.

[0069] In the embodiment of the present invention, the detection device 108 may be a camera, a camera or an infrared ray device, etc. The detection device may be deployed in the application environment, and may photograph the target area from multiple different angles to obtain a viewing angle image at each angle.

[0070] The determining module comprises:

[0071] An analysis submodule, configured to perform object analysis on each of the view angle images to determine an object region where an object exists at each view angle;

[0072] The mapping submodule is used to map the object area where the object exists in each viewing angle to the reference image, and determine the object area contained in the reference image.

[0073] In the embodiment of the present invention, the determination module includes an analysis submodule and a mapping submodule, wherein the analysis submodule performs object analysis on each viewing angle image to determine the object region where the object exists in the viewing angle image at each viewing angle.

[0074] The mapping submodule determines the entire object area included in the reference image by mapping the object area where the object exists at the current angle with the reference image.

[0075] The interception module comprises:

[0076] A judging submodule, used for judging whether the object region contained in the reference image is a continuous region;

[0077] The interception submodule is used to intercept the images corresponding to the discrete object regions contained in the reference image respectively if the object region contained in the reference image is not a continuous region, and use the intercepted images as the images to be identified.

[0078] In an embodiment of the present invention, the interception module includes a judgment submodule and a interception submodule, wherein the judgment submodule judges whether the object area contained in the reference image is a continuous area. In an embodiment of the present invention, the objects contained in the reference image may be connected together in the same area, or may be dispersedly arranged. The judgment submodule determines whether the part where the object area exists is a continuous area.

[0079] The interception submodule intercepts the image corresponding to the area containing the object. If the object is in the same area, one image is intercepted as the image to be identified. If it is a scattered area, an image is intercepted for each area and each intercepted image is used as the image to be identified.

[0080] Optionally, in the embodiment of the present invention, the captured images may be integrated as the image to be recognized.

[0081] The enhancement module comprises:

[0082] The enhancement submodule is used to perform enhancement processing on the image to be identified, wherein the enhancement processing at least includes any one or a combination of any several of grayscale, binarization, denoising and contrast enhancement.

[0083] In the embodiment of the present invention, the enhancement module includes an enhancement submodule, which can perform enhancement processing on the image to be identified and has the ability to perform operations such as grayscale, binarization, denoising and contrast enhancement.

[0084] In the embodiment of the present invention, the enhancement submodule needs to perform any one or any combination of grayscale, binarization, denoising and contrast enhancement on the image to be identified.

[0085] The system provided by the embodiment of the present invention further includes:

[0086] The display module 109 is used to display to the user the identification result of whether the target object output by the deep learning model is a hazardous chemical, and when the target object is a hazardous chemical, the chemical type of the hazardous chemical of the target object.

[0087] In the embodiment of the present invention, the system is provided with a display module 109 for displaying corresponding identification information to the user.

[0088] The training module comprises:

[0089] The cleaning submodule is used to clean the multiple test images to remove blurred images and repeated images in the multiple test images.

[0090] In the embodiment of the present invention, a cleaning submodule is provided in the training module, which is used to clean the multiple test images and delete the blurred images and duplicate images in the multiple test images.

[0091] The system provided by the embodiment of the present invention further includes:

[0092] The model testing module 110 is used to test the established deep learning model to determine the processing performance of the deep learning model.

[0093] The model optimization module 111 is used to optimize the established deep learning model to adjust the model structure and hyperparameters of the deep learning model.

[0094] The model deployment interface 112 is used to connect the deep learning model to the cloud server or edge device to complete the deployment of the deep learning model in the cloud server or edge device.

[0095] In summary, the embodiment of the present invention provides a hazardous chemical identification system based on deep learning and automatic image recognition. In practical applications, the system can inherit multiple target detection algorithms, such as the YOLOv5 target detection algorithm, etc. The application of the hazardous chemical identification system can realize efficient and accurate identification of various hazardous chemicals, which significantly improves the detection efficiency and accuracy of hazardous chemicals.

[0096] The hazardous chemical identification system based on deep learning and automatic image recognition provided by the embodiment of the present invention can use a hazardous chemical image dataset containing labels for model training in the dataset preparation stage to ensure that the model can identify various types of hazardous chemicals. The dataset can cover a wide range of chemical categories, such as flammables, explosives, toxic substances, etc., and perform strict data labeling and cleaning.

[0097] During the model training process, a variety of model algorithms can be used. For example, a pre-trained YOLOv5 model can be integrated and fine-tuned using a hazardous chemicals dataset to adapt to specific recognition tasks. The YOLOv5 model uses a lightweight CSPDarknet53 backbone network and a PANet feature fusion method, which can effectively extract feature information from images and improve detection accuracy and speed. The system provided by an embodiment of the present invention can select a specific model algorithm to train a deep learning model according to the application scenario that actually needs to be identified.

[0098] In the system provided by the embodiment of the present invention, the established deep learning model can be deployed in the inference engine to realize real-time or offline identification of hazardous chemicals. The inference engine supports the acquisition of image data from various security inspection equipment, surveillance cameras, etc., and quickly outputs the detection results, including information such as the category, location and confidence of hazardous chemicals.

[0099] The system provided by the embodiment of the present invention can provide an intuitive and friendly user interface, which is convenient for users to operate and view recognition results. The user interface supports multiple modes such as picture detection, video detection and camera real-time detection, and can be flexibly configured according to actual needs.

[0100] In actual application, if the above YOLOv5 is used to train the deep learning model, the following application examples can be provided:

[0101] We collected and organized an image dataset of 10 types of common hazardous chemicals, including 1,288 images in total, and annotated them in detail. The annotation format is a txt file in YOLO format, which is convenient for model training.

[0102] The YOLOv5 model was trained using Python, OpenCV library, and PyTorch framework in combination with the above datasets. Data enhancement technology was used during the training process to increase the generalization ability of the model, and the model parameters were optimized through multiple iterations until the ideal recognition effect was achieved.

[0103] Deploy the trained model in the inference engine and integrate it into the security inspection equipment or monitoring system. Verify the recognition accuracy and real-time performance of the system through actual testing to ensure that it meets the application requirements.

[0104] Design and implement a Qt or Web-based user interface that supports real-time detection of images, videos, and cameras. Users can select detection modes, upload images or video files, and view the recognition results output by the system through the interface.

[0105] The hazardous chemical identification system based on deep learning and automatic image recognition and classification provided by an embodiment of the present invention can collect a large amount of hazardous chemical image data before model training. These images should cover hazardous chemicals of different types and forms (such as liquids, solids, and gas containers).

[0106] Data sources may include laboratories, industrial production sites, warehouses, transport vehicles, etc.

[0107] The collected images are cleaned to remove blurry, duplicate or irrelevant images.

[0108] Perform image enhancement processing, such as grayscale, binarization, denoising, contrast enhancement, etc., to improve image quality and recognition accuracy.

[0109] Annotate images and add labels to each image to indicate the types of hazardous chemicals it contains.

[0110] Deep learning techniques, such as convolutional neural networks (CNNs), are used to automatically extract features related to hazardous chemicals from pre-processed images.

[0111] These features may include shape, color, texture, edges, etc.

[0112] Select a suitable deep learning model (such as YOLO, SSD, Faster R-CNN, etc.) for training.

[0113] Use labeled image data as a training set, and continuously adjust model parameters through forward propagation and back propagation algorithms to optimize the recognition performance of the model.

[0114] During the training process, cross-validation, regularization, dropout and other techniques can be used to prevent overfitting.

[0115] Tune the trained model, including adjusting the model structure, hyperparameters, etc., to improve the accuracy and efficiency of the model.

[0116] Methods such as ensemble learning and transfer learning can be used to further improve model performance.

[0117] The model is tested using an independent test set to evaluate its performance and accuracy in real-world applications.

[0118] The test set should include images of hazardous chemicals of different types and forms to comprehensively test the generalization ability of the model.

[0119] Apply the trained model to actual scenarios, such as chemical production sites, warehouses, transport vehicles, etc.

[0120] Images are captured in real time through image acquisition devices such as cameras and input into the model for recognition.

[0121] Deploy the model to edge devices (such as cameras, smart terminals, etc.) or cloud servers.

[0122] Integrate the model with existing monitoring systems, alarm systems, etc. to achieve real-time detection and early warning of hazardous chemicals.

[0123] Continuously evaluate the performance of the model in actual applications, including indicators such as recognition accuracy and response time.

[0124] The model is updated and optimized regularly to adapt to new application scenarios and needs.

[0125] New image data is continuously collected and used for retraining and tuning the model.

[0126] Focus on new technologies and methods in the field of deep learning, and try to apply them to the field of hazardous chemical identification.

[0127] For the specific working process of each module and sub-module in the hazardous chemical identification system based on deep learning and automatic image recognition and classification disclosed in the above embodiment of the present invention, please refer to the corresponding content of the hazardous chemical identification system based on deep learning and automatic image recognition and classification disclosed in the above embodiment of the present invention, which will not be repeated here.

[0128] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0129] Those skilled in the art may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, computer software, or a combination of both.

[0130] In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0131] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dangerous chemical identification system based on deep learning and automatic image recognition and classification, characterized in that: include: A receiving module, used to receive images of various viewing angles obtained by the detection device shooting the target area at multiple angles; A selection module, used for selecting a viewing angle image taken at a normal viewing angle as a reference image from among the viewing angle images; A determination module, used for determining an object area included in the reference image according to each of the view images; A capture module, used to capture the image where the object area is located as the image to be recognized; An enhancement module, used to perform enhancement processing on the image to be identified to obtain a target image corresponding to the image to be identified; A recognition module, used to input the target image into an established deep learning model, determine the target objects contained in the target image and the object form of each target object through processing by the deep learning model, and extract the object features of each target object according to the object form of each target object, identify whether the target object is a hazardous chemical based on the object features of each target object, and output the chemical type of the hazardous chemical to which the target object belongs when the target object is identified as a hazardous chemical; A training module, used to determine a plurality of test images, wherein the plurality of test images include images of various types of hazardous chemicals; Performing enhancement processing on the multiple test images to obtain multiple sample images; Determine each object contained in each of the sample images, and add an attribute label to each object in each of the sample images; Extracting object features of each object in each of the sample images, wherein the object features at least include the shape, color, texture and edge of the object; An initial learning model is determined, and each object feature of each object in each of the sample images is used as input, and each attribute label of each object in each of the sample images is used as output. The initial learning model is trained until the convergence function of the initial learning model is less than a preset threshold, and the training of the initial learning model is completed to establish the deep learning model; each of the attribute labels is used to characterize whether the object corresponding to it is a hazardous chemical, and when the object is a hazardous chemical, the chemical type of the hazardous chemical to which the object belongs.

2. The dangerous chemical identification system based on deep learning and automatic image recognition and classification according to claim 1 is characterized in that: Also includes: The detection device is used to photograph the target area from multiple angles to obtain a viewing angle image at each angle.

3. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: The determining module comprises: An analysis submodule, configured to perform object analysis on each of the view angle images to determine an object region where an object exists at each view angle; The mapping submodule is used to map the object area where the object exists in each viewing angle to the reference image, and determine the object area contained in the reference image.

4. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: The interception module comprises: A judging submodule, used for judging whether the object region contained in the reference image is a continuous region; The interception submodule is used to intercept the images corresponding to the discrete object regions contained in the reference image respectively if the object region contained in the reference image is not a continuous region, and use the intercepted images as the images to be identified.

5. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: The enhancement module comprises: The enhancement submodule is used to perform enhancement processing on the image to be identified, wherein the enhancement processing at least includes any one or a combination of any several of grayscale, binarization, denoising and contrast enhancement.

6. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: Also includes: The display module is used to display to the user the identification result of whether the target object output by the deep learning model is a hazardous chemical, and when the target object is a hazardous chemical, the chemical type of the hazardous chemical of the target object.

7. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: The training module comprises: The cleaning submodule is used to clean the multiple test images to remove blurred images and repeated images in the multiple test images.

8. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: Also includes: The model testing module is used to test the established deep learning model to determine the processing performance of the deep learning model.

9. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: Also includes: The model optimization module is used to optimize the established deep learning model to adjust the model structure and hyperparameters of the deep learning model.

10. The dangerous chemical identification system based on deep learning and automatic image recognition according to claim 1 is characterized in that: Also includes: The model deployment interface is used to connect the deep learning model to the cloud server or edge device to complete the deployment of the deep learning model in the cloud server or edge device.

Images