AI visual detection method and system for carrying liquid water when people pass

Abstract

The invention relates to the technical field of artificial intelligence and computer vision, and provides an AI visual detection method and system for liquid water carried by people during passing, and the method comprises the steps: constructing an AI visual detection model based on a convolutional neural network; acquiring a target image of personnel passing in real time and performing image preprocessing; inputting the target image into a trained AI visual detection model, and extracting visual features of a person carrying object through a feature extractor; and the classifier performs matching according to the visual features and a feature database established during pre-training to judge whether the articles carried by the personnel in the target image are liquid water or not, and when the classification result is liquid water, an alarm is automatically triggered. By realizing real-time and automatic liquid detection, the detection efficiency is improved, the cost depending on manpower and expensive equipment is reduced, the risk of false alarm and omission is reduced, various detection environments are flexibly coped with, an economical and efficient solution is provided, and the method is suitable for being applied to public places needing rapid mass people stream processing and strict safety inspection.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and computer vision technology, and in particular to an AI vision detection method and system for detecting liquid water carried by people when they pass by. Background Art

[0002] With the increasing social security requirements, especially in public places such as airports, stations, stadiums, large conferences, and large events, it has become a top priority to ensure that the items carried by people do not pose a threat to public safety. In these places, the carrying and transportation of liquids, especially liquids such as alcohol, may pose certain risks to safety. Especially in some special cases, the carrying of liquids may be a potential source of danger. Therefore, as an important part of the security inspection system, the accuracy and efficiency of liquid detection have become the key to improving the level of security management.

[0003] Currently, the most common methods for detecting liquid water include X-ray scanning and manual inspection. These methods can complete the task of detecting liquids to a certain extent, but these methods are inefficient and often lead to long queues, especially during peak hours, which seriously affects traffic efficiency and customer experience. Secondly, these methods often rely on the experience and judgment of security personnel and are easily affected by human factors, resulting in frequent false alarms and missed alarms. In addition, X-ray technology can detect liquid items more effectively, but its equipment is expensive, the operation is complicated, and professional training of staff is required. More importantly, X-ray radiation may have a certain impact on human health, especially when it needs to be used for a long time.

[0004] Therefore, there is an urgent need for a more efficient, safer and more reliable technology to quickly and accurately identify and distinguish liquid substances, significantly improve the detection speed and accuracy, reduce the reliance on manual operation, reduce the false alarm rate, and completely avoid the impact on human health. Summary of the invention

[0005] In view of the above problems, the present invention proposes an AI visual detection method and system for liquid water carried by people when passing by, so as to realize real-time and accurate detection of liquids carried by people. The above invention object of the present invention is achieved through the following technical solutions: The present invention provides an AI visual detection method for liquid water carried by people when passing by, comprising: Build an AI visual detection model based on convolutional neural networks to extract and learn features from target images, and train the model by labeling water and non-water image datasets; Collect target images of people passing by in real time and pre-process the images to conform to the input format of the trained AI visual detection model; The pre-processed target image is input into the trained AI visual detection model, and the visual features of the objects carried by the person are extracted through the feature extractor; The classifier matches the visual features with the feature database established during pre-training to determine whether the items carried by the person in the target image are liquid water. If the classification result is liquid water, an alarm is automatically triggered; otherwise, the person is allowed to pass.

[0006] Furthermore, a convolutional neural network-based AI visual detection model is constructed to extract and learn features from target images, including: Construct an AI visual detection model, which includes a feature extractor and a classifier. The feature extractor includes three convolutional layers and corresponding pooling layers, which are used to extract visual features of the target image by convolution downsampling; the classifier includes a flattening layer, a fully connected layer with Dropout, and a fully connected layer that outputs binary classification probabilities, which are used to perform binary classification operations on the target image and output the corresponding classification results.

[0007] Furthermore, the model is trained by labeling water and non-water image datasets, including, Collect and mark water and non-water image datasets, and perform image preprocessing on target images in the image dataset; Based on the constructed AI visual inspection model, start training the model by calling the training function, passing in the training data generator, number of steps per round, verification data generator, number of verification steps, and number of training rounds; and evaluate the AI ​​visual inspection model based on the loss value and / or accuracy through the verification data generator.

[0008] Furthermore, according to the AI ​​visual detection method of claim 1, it is characterized in that real-time acquisition of target images of people passing by and image preprocessing are performed, further comprising: The target image is loaded and converted to grayscale format, the horizontal and vertical gradients are calculated using edge detection algorithm, and the image is converted to binary format through threshold processing to highlight visual features.

[0009] Furthermore, the preprocessed target image is input into the trained AI visual detection model, and the visual features of the objects carried by the person are extracted through the feature extractor, including: The feature extractor consists of three convolutional layers and corresponding pooling layers. The convolution kernels of the convolutional layers are 32, 64, and 128. Multi-layer visual features are extracted from the target image through layer-by-layer convolution and downsampling.

[0010] Furthermore, the classifier determines whether the object carried by the person in the target image is liquid water by matching the visual features with the feature database established during pre-training, including: The flattening layer in the classifier converts multi-layer visual features into a one-dimensional array, and enhances the generalization ability of the AI ​​visual detection model through a fully connected layer including a Dropout operation, and outputs the classification results through a sigmoid activation function to achieve a binary classification result of whether the target in the target image is liquid water.

[0011] Furthermore, based on the constructed AI visual detection model, the model training is started by calling the training function, passing in the training data generator, the number of iteration steps per round, the verification data generator, the number of verification steps, and the number of training cycles, including: Initialize the training data generator and the verification data generator, where the training data generator loads data augmentation from the specified training data directory, including rotation, scaling, and horizontal flipping, and the verification data generator loads the target image from the verification data directory and performs image preprocessing; Setting training parameters is used to adjust the AI ​​visual inspection model. Training parameters include the number of iterations per round, the number of verification steps, and the number of training cycles.

[0012] Furthermore, the AI ​​visual inspection model is evaluated based on the loss value and / or accuracy through the validation data generator, including: During the training process, the loss value and accuracy at the end of each training cycle are output in real time to monitor the progress of model training. After all training cycles are completed, the loss value and accuracy of the last verification cycle are output as the final evaluation indicators of model performance.

[0013] Furthermore, based on the classification results of the AI ​​visual detection model, an alarm is automatically triggered when the classification result is liquid, otherwise people are allowed to pass through without obstacles, including: After the AI ​​visual detection model determines that the objects carried by the person in the target image are liquid water, it automatically activates the preset alarm mechanism, including sound or light signal alarms.

[0014] Based on the same inventive concept, the present invention also provides an AI visual detection system for people carrying liquid water when passing by, using the above-mentioned AI visual detection method, including: The data acquisition module is used to collect target images of people passing by in real time and perform image preprocessing to conform to the input format of the trained AI visual detection model; The data processing module is used to input the preprocessed target image into the trained AI visual detection model, and extract the visual features of the person's carrying objects through the feature extractor; wherein the AI ​​visual detection model includes a feature extractor and a classifier, wherein the feature extractor includes three convolutional layers and corresponding pooling layers, which are used to extract the visual features of the target image by convolution downsampling; the classifier includes a flattening layer, a fully connected layer with Dropout, and a fully connected layer that outputs binary classification probabilities, which is used to perform binary classification operations on the target image and output the corresponding classification results; The result output module is used for the classifier to match the visual features with the feature database established during pre-training to determine whether the items carried by the person in the target image are liquid water. If the classification result is liquid water, an alarm is automatically triggered; otherwise, the person is allowed to pass.

[0015] Compared with the prior art, the present invention has at least one of the following beneficial effects: By achieving real-time, automated liquid detection, the detection efficiency is improved and the cost of relying on manpower and expensive equipment is reduced, while ensuring a high degree of safety. At the same time, this method reduces the risk of false alarms and omissions, and through its high adaptability and scalability, it can flexibly respond to various detection environments, providing a cost-effective solution, suitable for public places that require rapid large-scale crowd processing and strict security checks.

[0016] (1) The AI ​​visual inspection technology of the present invention realizes real-time detection of liquids carried by people through an automated processing flow, significantly improving the detection speed and process efficiency. This reduces manual queuing and waiting time, speeds up the flow of people, and is particularly suitable for public places with high passenger flow density.

[0017] (2) Compared with traditional liquid detection methods (such as manual inspection or X-ray scanning), AI visual inspection systems reduce reliance on human resources and the need for expensive equipment, thereby significantly reducing operating costs. In addition, maintenance and update costs are relatively low, making the long-term return on investment more significant.

[0018] (3) It can accurately and quickly identify whether the items being carried are liquids, effectively preventing potentially dangerous liquids from entering safety-sensitive areas, reducing the risk of false alarms and omissions, and improving the overall safety management level. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flowchart of the steps of AI visual detection in an embodiment of the present invention; Figure 2 Flow chart of the AI ​​visual inspection process in an embodiment of the present invention; Figure 3 This is a system architecture diagram for the deployment of the AI ​​visual detection model in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0022] First embodiment In large public places such as airports and stations, facing the increasing flow of people, traditional security detection methods (including manual inspection and basic mechanical scanning) face inefficiencies and easy congestion, especially when it is necessary to identify non-metallic objects such as liquid substances. These traditional methods often rely on a lot of manual operations, which not only increases labor costs, but also may cause safety hazards due to human negligence. To solve these problems, the inventors proposed an intelligent AI visual inspection system based on deep learning and computer vision technology. The system uses convolutional neural networks (CNN) for real-time image analysis, aiming to improve the speed and accuracy of detection, reduce dependence on manual inspection, and effectively cope with the challenges brought by high traffic.

[0023] Based on this, the inventor proposed a new technical solution, and developed an AI visual inspection system by combining deep learning algorithms and convolutional neural network (CNN) technology. The system can automatically detect liquid items of individuals passing through security checks in real time, which not only significantly improves the speed and accuracy of detection, but also greatly reduces operating costs and environmental impact. Through intelligent image analysis technology, the system can effectively identify and classify liquid items carried by individuals, ensure the strict implementation of security measures, and ensure the smooth passage of people. This innovative solution is expected to bring revolutionary changes in the field of security inspections. The specific implementation methods are as follows: The present invention provides an AI visual detection method for liquid water carried by people when passing by, such as Figure 1 As shown, including, Build an AI visual detection model based on convolutional neural networks to extract and learn features from target images, and train the model by labeling water and non-water image datasets; Collect target images of people passing by in real time and pre-process the images to conform to the input format of the trained AI visual detection model; The pre-processed target image is input into the trained AI visual detection model, and the visual features of the objects carried by the person are extracted through the feature extractor; The classifier matches the visual features with the feature database established during pre-training to determine whether the items carried by the person in the target image are liquid water. If the classification result is liquid water, an alarm is automatically triggered; otherwise, the person is allowed to pass.

[0024] Furthermore, a convolutional neural network-based AI visual detection model is constructed to extract and learn features from target images, including: Construct an AI visual detection model, which includes a feature extractor and a classifier. The feature extractor includes three convolutional layers and corresponding pooling layers, which are used to extract visual features of the target image by convolution downsampling; the classifier includes a flattening layer, a fully connected layer with Dropout, and a fully connected layer that outputs binary classification probabilities, which are used to perform binary classification operations on the target image and output the corresponding classification results.

[0025] Furthermore, the model is trained by labeling water and non-water image datasets, including, Collect and mark water and non-water image datasets, and perform image preprocessing on target images in the image dataset; Based on the constructed AI visual inspection model, start training the model by calling the training function, passing in the training data generator, number of steps per round, verification data generator, number of verification steps, and number of training rounds; and evaluate the AI ​​visual inspection model based on the loss value and / or accuracy through the verification data generator.

[0026] The AI ​​visual detection model uses deep learning technology to identify water, especially convolutional neural networks to extract features from images and identify objects in the images. The following detailed steps are explained: (1) Data preparation: Collect and prepare a dataset of images labeled as “water” and “non-water”. This dataset will be used as the basis for training and validating the model.

[0027] (2) Data preprocessing: Image data is preprocessed through ImageDataGenerator to convert it into a format suitable for model training. This includes scaling pixel values ​​to between 0 and 1 (normalization) and applying a series of data augmentation techniques such as rotation, scaling, horizontal flipping, etc. to improve the model's generalization ability to new and unseen data.

[0028] (3) Build an AI visual detection model, using the Sequential model of the Keras framework to stack different layers and build a CNN. The model includes: convolutional layer (Conv2D), used to extract features (such as edges, textures, etc.) in the image; pooling layer (MaxPooling2D), used to reduce the spatial dimension of the features, reduce the amount of calculation, while maintaining important features; flattening layer (Flatten), converting the two-dimensional feature map into one dimension for use in the fully connected layer; fully connected layer (Dense), used to learn feature combinations; Dropout layer, randomly discarding a certain proportion of nodes during training to avoid overfitting; output layer, using the sigmoid activation function to predict whether the image is water or non-water.

[0029] (4) Compile the model, use binary cross entropy as the loss function, and use the Adam optimizer for parameter optimization, while tracking accuracy as the performance metric.

[0030] (5) Train the model. The model is trained using the training data generator, specifying the number of steps and validation data for each epoch. The model is trained through multiple epochs, each epoch iterating on the training set and evaluating the performance on the validation set.

[0031] (6) Evaluate the model. After training, save the model and perform a final evaluation on an independent validation set to check its generalization ability. Output the model's loss value and accuracy to evaluate its performance.

[0032] Furthermore, according to the AI ​​visual detection method of claim 1, it is characterized in that real-time acquisition of target images of people passing by and image preprocessing are performed, further comprising: The target image is loaded and converted to grayscale format, the horizontal and vertical gradients are calculated using edge detection algorithm, and the image is converted to binary format through threshold processing to highlight visual features.

[0033] Furthermore, the preprocessed target image is input into the trained AI visual detection model, and the visual features of the objects carried by the person are extracted through the feature extractor, including: The feature extractor consists of three convolutional layers and corresponding pooling layers. The convolution kernels of the convolutional layers are 32, 64, and 128. Multi-layer visual features are extracted from the target image through layer-by-layer convolution and downsampling.

[0034] Furthermore, the classifier determines whether the object carried by the person in the target image is liquid water by matching the visual features with the feature database established during pre-training, including: The flattening layer in the classifier converts multi-layer visual features into a one-dimensional array, and enhances the generalization ability of the AI ​​visual detection model through a fully connected layer including a Dropout operation, and outputs the classification results through a sigmoid activation function to achieve a binary classification result of whether the target in the target image is liquid water.

[0035] Furthermore, based on the constructed AI visual detection model, the model training is started by calling the training function, passing in the training data generator, the number of iteration steps per round, the verification data generator, the number of verification steps, and the number of training cycles, including: Initialize the training data generator and the verification data generator, where the training data generator loads data augmentation from the specified training data directory, including rotation, scaling, and horizontal flipping, and the verification data generator loads the target image from the verification data directory and performs image preprocessing; Setting training parameters is used to adjust the AI ​​visual inspection model. Training parameters include the number of iterations per round, the number of verification steps, and the number of training cycles.

[0036] Furthermore, the AI ​​visual inspection model is evaluated based on the loss value and / or accuracy through the validation data generator, including: During the training process, the loss value and accuracy at the end of each training cycle are output in real time to monitor the progress of model training. After all training cycles are completed, the loss value and accuracy of the last verification cycle are output as the final evaluation indicators of model performance.

[0037] Furthermore, based on the classification results of the AI ​​visual detection model, an alarm is automatically triggered when the classification result is liquid, otherwise people are allowed to pass through without obstacles, including: After the AI ​​visual detection model determines that the objects carried by the person in the target image are liquid water, it automatically activates the preset alarm mechanism, including sound or light signal alarms.

[0038] The specific implementation steps of the present invention are as follows: Figure 2 As shown, (1) Data collection: using high-definition cameras and other image acquisition equipment to collect images or video data of people passing by.

[0039] (2) Data processing: input the collected image or video data into the deep learning algorithm for processing to identify the items carried by the person.

[0040] (3) Liquid detection: Through the trained AI model, namely the AI ​​visual detection model, the identified objects are further analyzed to determine whether they are liquids. The AI ​​model can be trained and optimized through a large amount of training data and labels to improve the accuracy and stability of detection.

[0041] (4) Alarm processing: If it is determined to be liquid, an alarm signal is issued and relevant information (such as the image of the person, the type of liquid carried, etc.) is sent to the safety manager for processing. The safety manager can take corresponding measures according to the actual situation, such as intercepting the person, further inspection, etc.

[0042] Second embodiment Based on the same inventive concept, the present invention also provides an AI visual detection system for people carrying liquid water when passing by, using the above-mentioned AI visual detection method, including: The data acquisition module is used to collect target images of people passing by in real time and perform image preprocessing to conform to the input format of the trained AI visual detection model; The data processing module is used to input the preprocessed target image into the trained AI visual detection model, and extract the visual features of the person's carrying objects through the feature extractor; wherein the AI ​​visual detection model includes a feature extractor and a classifier, wherein the feature extractor includes three convolutional layers and corresponding pooling layers, which are used to extract the visual features of the target image by convolution downsampling; the classifier includes a flattening layer, a fully connected layer with Dropout, and a fully connected layer that outputs binary classification probabilities, which is used to perform binary classification operations on the target image and output the corresponding classification results; The result output module is used for the classifier to match the visual features with the feature database established during pre-training to determine whether the items carried by the person in the target image are liquid water. If the classification result is liquid water, an alarm is automatically triggered; otherwise, the person is allowed to pass.

[0043] like Figure 3 As shown, the AI ​​visual recognition system is deployed in high-traffic public areas such as airports and stations to detect whether people are carrying liquid water in real time. The architecture is divided into four main layers: user interaction layer, front-end collection layer, back-end processing layer, and alarm and response layer.

[0044] User interaction layer, which provides a user-friendly interface that allows system administrators to configure settings, monitor video streams passing through detection areas in real time, and query historical alarm records. Through this interface, users can easily manage and monitor the operating status of the entire system.

[0045] Front-end acquisition layer, in this layer, high-definition cameras are responsible for capturing images of individuals passing by and sending video data to the back-end processing layer in real time through the data transmission module. These cameras are deployed in key locations to ensure that sufficient image data can be captured from all angles.

[0046] At the back-end processing layer, the received data is first processed by the data receiving module and then enters the preprocessing module to perform preprocessing operations such as format conversion and quality enhancement on the image. The processed image is sent to the detection judgment module, which uses a pre-trained deep learning AI intelligent algorithm to determine whether the image contains liquid water. All detection data is securely stored by the data storage module for future query and analysis.

[0047] Alarm and response layer,When the detection and judgment module confirms the presence of liquid water in the image, the alarm module will trigger an alarm signal to notify the security management personnel to take corresponding measures. The response module coordinates subsequent actions, including but not limited to intercepting people suspected of carrying prohibited items.

[0048] The system is designed with efficiency, safety and low-cost operation in mind, and is suitable for high-risk environments that require fast and accurate detection. By combining advanced image processing technology with deep learning algorithms, the present invention can provide an automated and intelligent solution, greatly improving the efficiency and accuracy of security detection.

[0049] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

[0050] It should be noted that the above embodiments can be freely combined as needed. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention.

Claims

1. An AI visual detection method for liquid water carried by people passing by, characterized in that: include, Build an AI visual detection model based on convolutional neural networks to extract and learn features from target images, and train the model by labeling water and non-water image datasets; Collect the target image of the person passing by in real time and perform image preprocessing to conform to the input format of the trained AI visual detection model; Input the preprocessed target image into the trained AI visual detection model, and extract the visual features of the person's belongings through a feature extractor; The classifier matches the visual features with the feature database established during pre-training to determine whether the item carried by the person in the target image is liquid water. If the classification result is liquid water, an alarm is automatically triggered; otherwise, the person is allowed to pass.

2. The AI ​​visual inspection method according to claim 1, characterized in that: Build an AI visual detection model based on convolutional neural network to extract and learn features from target images, further including, Construct the AI ​​visual detection model, which includes the feature extractor and the classifier, wherein the feature extractor includes three convolutional layers and corresponding pooling layers, which are used to extract the visual features of the target image obtained by convolution downsampling; the classifier includes a flattening layer, a fully connected layer with Dropout, and a fully connected layer that outputs binary classification probabilities, which is used to perform a binary classification operation on the target image and output the corresponding classification result.

3. The AI ​​visual inspection method according to claim 2, characterized in that: Model training is performed by labeling water and non-water image datasets; further comprising, Collect and mark the image dataset as water and non-water, and perform image preprocessing on the target image in the image dataset; Based on the constructed AI visual inspection model, the model training is started by calling the training function, passing in the training data generator, the number of steps per round, the verification data generator, the number of verification steps and the number of training rounds; and the AI ​​visual inspection model is evaluated based on the loss value and / or accuracy rate through the verification data generator.

4. The AI ​​visual inspection method according to claim 1, characterized in that: The AI ​​visual detection method according to claim 1 is characterized in that the target image of a person passing by is collected in real time and image preprocessing is performed, further comprising: The target image is loaded and converted into a grayscale format, the horizontal and vertical gradients are calculated using an edge detection algorithm, and the image is converted into a binary format through threshold processing to highlight the visual features.

5. The AI ​​visual inspection method according to claim 2, characterized in that: Inputting the preprocessed target image into the trained AI visual detection model, extracting the visual features of the person's belongings through a feature extractor, further comprising: The feature extractor includes three convolutional layers and corresponding pooling layers, the convolution kernels of the convolutional layers are 32, 64 and 128, and multi-layer visual features are extracted from the target image through layer-by-layer convolution and downsampling.

6. The AI ​​visual inspection method according to claim 5, characterized in that: The classifier determines whether the item carried by the person in the target image is liquid water by matching the visual feature with a feature database established during pre-training, further comprising: The flattening layer in the classifier converts the multi-layer visual features into a one-dimensional array, and enhances the generalization ability of the AI ​​visual detection model through the fully connected layer including the Dropout operation, and outputs the classification result through the sigmoid activation function to achieve a binary classification result of whether the target in the target image is the liquid water.

7. The AI ​​visual inspection method according to claim 3, characterized in that: Based on the constructed AI visual inspection model, by calling the training function, passing in the training data generator, the number of iteration steps per round, the verification data generator, the number of verification steps and the number of training cycles parameters, the model training is started, further including: Initialize the training data generator and the verification data generator, wherein the training data generator loads the data enhancement from the specified training data directory, including rotation, scaling and horizontal flipping, and the verification data generator loads the target image from the verification data directory and performs the image preprocessing; The training parameters are set to adjust the AI ​​visual detection model, and the training parameters include the number of iteration steps per round, the number of verification steps, and the number of training cycles.

8. The AI ​​visual inspection method according to claim 4, characterized in that: Evaluating the AI ​​visual inspection model based on the loss value and / or accuracy by the verification data generator further includes: During the training process, the loss value and accuracy at the end of each training cycle are output in real time to monitor the progress of model training. After all training cycles are completed, the loss value and accuracy of the last verification cycle are output as the final evaluation indicators of model performance.

9. The AI ​​visual inspection method according to claim 1, characterized in that: Based on the classification result of the AI ​​visual detection model, when the classification result is the liquid, an alarm is automatically triggered, otherwise the person is allowed to pass through without obstacles, further comprising: After the AI ​​visual detection model determines that the object carried by the person in the target image is the liquid water, it automatically activates a preset alarm mechanism, including a sound or light signal alarm.

10. An AI visual detection system for liquid water carried by people when passing by, using the AI ​​visual detection method according to any one of claims 1 to 9, characterized in that: include, A data acquisition module, used to collect the target images of people passing by in real time and perform image preprocessing to conform to the input format of the trained AI visual detection model; A data processing module, used to input the pre-processed target image into the trained AI visual detection model, and extract the visual features of the person's carrying objects through a feature extractor; wherein the AI ​​visual detection model includes the feature extractor and the classifier, wherein the feature extractor includes three convolutional layers and corresponding pooling layers, which are used to extract the visual features obtained by convolution downsampling the target image; the classifier includes a flattening layer, a fully connected layer with Dropout, and a fully connected layer that outputs binary classification probabilities, which is used to perform a binary classification operation on the target image and output the corresponding classification result; The result output module is used for the classifier to match the visual features with the feature database established during pre-training to determine whether the item carried by the person in the target image is liquid water. When the classification result is liquid water, an alarm is automatically triggered; otherwise, the person is allowed to pass.

Images