An image recognition model and training method for illegal object detection, an illegal object detection method and device, and an autonomous mobile robot

Abstract

The present invention provides an image recognition model and training method for detecting illegal items, a method and device for detecting illegal items, and an autonomous mobile robot. By introducing a neural network model for automatic recognition, the intelligent level of security monitoring is greatly improved, which reduces the dependence on human resources and reduces long-term operating costs; through advanced recognition algorithms, the possibility of missed detection is reduced. In addition, automated scanning reduces manual intervention and further improves the logistics processing speed.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to an image recognition model and training method for detecting illegal items, a method and device for detecting illegal items, and an autonomous mobile robot. Background Art

[0002] Traditional aviation logistics security checks rely on manual or semi-automated scanning equipment, which have problems such as low efficiency, easy to miss inspections, and high costs. For example, the average amount of goods processed per person per hour for manual sorting is only 50-80 pieces, which cannot cope with the huge cargo flow. In addition, the traditional automated sorting method relies on barcode recognition, which is easily affected by the quality of the barcode and has a low reading success rate, resulting in cargo backlogs and delays, which seriously affects the flight punctuality rate. In addition, manual operation and traditional identification methods result in an information entry error rate of up to 15%-20%, and an omission rate of about 10%-15%. Inaccurate and incomplete cargo information makes tracking and management extremely difficult, and customers often face serious problems such as lost and mis-shipped goods, which triggers customer complaints and trust crises. In addition, during peak freight hours, the existing sorting system is completely unable to meet the demand, and the hourly processing capacity is far lower than the actual amount of goods that need to be processed. This leads to a large amount of cargo being stranded and the sorting site being in chaos, further exacerbating flight delays and customer dissatisfaction. Finally, the traditional sorting method requires a lot of manpower investment and high labor costs. At the same time, the additional transportation, storage and handling costs caused by sorting errors and delays continue to rise, seriously affecting the economic benefits and market competitiveness of enterprises. Therefore, an efficient and accurate logistics safety monitoring method is needed. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides an image recognition model and training method for detecting illegal items, a method and device for detecting illegal items, and an autonomous mobile robot, which mainly solve the problem of the lack of an efficient and accurate logistics safety monitoring method in the prior art.

[0004] The purpose of the present invention is achieved through the following solutions:

[0005] In a first aspect, according to an embodiment of the present invention, a training method for an image recognition model for illegal item detection is provided, comprising the following steps: S1, using a pre-trained image recognition model to initialize a region proposal network in a current image recognition model; S2, constructing a training data set, wherein the training data set includes images containing illegal items or non-illegal items, each image including an item category and a position as a label, wherein illegal items are items of a specified category; S3, training the current image recognition model based on the training data set, and updating the model parameters based on a preset loss function, wherein only the network layers in the current image recognition model that are different from the pre-trained image recognition model are updated.

[0006] According to an embodiment of the present invention, data enhancement is performed on images during the training process.

[0007] According to an embodiment of the present invention, during the training process, the detection boxes generated by the region proposal network are overlapped and merged.

[0008] According to an embodiment of the present invention, overlapping merging adopts non-maximum suppression.

[0009] According to an embodiment of the present invention, the designated categories include knives, explosives, and drugs.

[0010] In a second aspect, according to another embodiment of the present invention, there is provided an image recognition model for detecting illegal items, which is obtained by using the training method described in the first aspect, wherein the image recognition model includes a region proposal network.

[0011] In a third aspect, according to another embodiment of the present invention, a method for detecting illegal items is provided, comprising the following steps: P1, acquiring an image of an object to be detected; P2, based on the acquired image, using the image recognition model as described in the second aspect to identify whether the object to be detected is an illegal item.

[0012] In a fourth aspect, according to another embodiment of the present invention, there is provided an apparatus for detecting illegal items, comprising: an X-ray unit or a millimeter wave scanning unit, for acquiring an image of an object to be detected; a detection module, configured with the image recognition model as described in the second aspect, and identifying whether the object to be detected is an illegal item based on the acquired image.

[0013] In a fifth aspect, according to another embodiment of the present invention, there is provided an autonomous mobile robot, which moves autonomously according to a preset program, and includes a device for detecting illegal items as described in the fourth aspect, so that illegal items can be detected during autonomous movement.

[0014] According to yet another embodiment of the present invention, the autonomous mobile robot further includes an alarm system for issuing a warning when an illegal item is detected.

[0015] Compared with the prior art, the present invention has the following beneficial effects: by introducing a neural network model for automatic identification and a corresponding automatic scanning device, the intelligent level of security monitoring is greatly improved. Among them, automatic scanning reduces manual intervention and improves the speed of logistics processing; reduces dependence on human resources and reduces long-term operating costs; and reduces the possibility of missed detection through advanced recognition algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The embodiments of the present invention are further described below with reference to the accompanying drawings:

[0017] Figure 1 The figure is a flowchart of a method for training an image recognition model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] As described in the background technology, the existing technology lacks an efficient and accurate logistics safety monitoring method. Figure 1 As shown, an embodiment of the present invention provides a training method for an image recognition model for detecting illegal items, comprising the following steps: S1, using a pre-trained image recognition model to initialize a region proposal network in a current image recognition model; S2, constructing a training data set, wherein the training data set includes images containing illegal items or non-illegal items, each image including an item category and a position as a label, wherein illegal items are items of a specified category; S3, training the current image recognition model based on the training data set, and updating the model parameters based on a preset loss function, wherein only the network layers in the current image recognition model that are different from the pre-trained image recognition model are updated. By introducing a neural network model for automatic recognition, the intelligent level of security monitoring is greatly improved, wherein the dependence on human resources is reduced and long-term operating costs are reduced; and the possibility of missed detection is reduced through advanced recognition algorithms. The preset loss function is cross entropy loss, etc.

[0020] In order to improve the generalization ability of the model, according to an embodiment of the present invention, data enhancement is performed on the image during the training process. Data enhancement techniques include rotation, flipping, scaling, etc.

[0021] In addition, in order to improve recognition accuracy, during the training process, the detection boxes generated by the region proposal network are overlapped and merged. According to an embodiment of the present invention, the overlapped merging adopts non-maximum suppression.

[0022] According to an embodiment of the present invention, the designated categories include knives, explosives, and drugs.

[0023] According to an embodiment of the present invention, the specific training method is as follows.

[0024] 1. Dataset construction:

[0025] Collect 10,000 images containing various illegal items (such as knives, explosives, drugs, etc.) and non-illegal items.

[0026] The image is scaled to 256x256 pixels and normalized, including scaling, cropping, and normalization.

[0027] Use the annotation tool to label the bounding box and category of the illegal objects in each image.

[0028] 2. Model training:

[0029] Use the Faster R-CNN model, fine-tuned based on a pre-trained ResNet-101 network (on a large dataset such as ImageNet).

[0030] The training is performed on the training set, using the stochastic gradient descent (SGD) optimizer with a learning rate of 0.001 for 10 epochs. For the same network layers in the Faster R-CNN model and the ResNet-101 network (e.g., the convolutional layers in the region proposal network), the learning rate is set to 0, that is, these layers are not updated.

[0031] 3. Model testing:

[0032] The model performance is evaluated on an independent test set consisting of 2000 images.

[0033] The model achieved 95% accuracy, with high confidence in detecting illegal items.

[0034] According to another embodiment of the present invention, there is provided an image recognition model for detecting illegal items, which is obtained by using the training method as described above, wherein the image recognition model includes a region proposal network.

[0035] According to another embodiment of the present invention, a method for detecting illegal items is provided, comprising the following steps: P1, acquiring an image of an object to be detected; P2, based on the acquired image, using the image recognition model as described above to identify whether the object to be detected is an illegal item.

[0036] According to another embodiment of the present invention, there is provided an illegal item detection device, comprising: an X-ray unit or a millimeter wave scanning unit, for acquiring an image of an object to be detected; a detection module, configured with the image recognition model as described above, and identifying whether the object to be detected is an illegal item based on the acquired image.

[0037] According to another embodiment of the present invention, there is provided an autonomous mobile robot, which moves autonomously according to a preset program, and includes a device for detecting illegal items as described above, so that illegal items can be detected during the autonomous movement.

[0038] According to yet another embodiment of the present invention, the autonomous mobile robot further includes an alarm system for issuing a warning when an illegal item is detected.

[0039] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A training method for an image recognition model for illegal item detection, characterized in that: The following steps are involved: S1. Use the pre-trained image recognition model to initialize the region proposal network in the current image recognition model; S2. Construct a training data set, wherein the training data set includes images containing illegal items or non-illegal items, each image includes an item category and a location as a label, wherein illegal items are items of a specified category; S3. Train the current image recognition model based on the training data set, and update the model parameters based on a preset loss function, wherein only the network layers in the current image recognition model that are different from the pre-trained image recognition model are updated.

2. The method for training an image recognition model for detecting illegal items according to claim 1, characterized in that: During the training process, data augmentation is performed on the images.

3. The method for training an image recognition model for detecting illegal items according to claim 1, characterized in that: During the training process, the detection boxes generated by the region proposal network are overlapped and merged.

4. The method for training an image recognition model for detecting illegal items according to claim 3, characterized in that: Overlap merging uses non-maximum suppression.

5. The method for training an image recognition model for detecting illegal items according to claim 1, characterized in that: Designated categories include knives, explosives and drugs.

6. An image recognition model for illegal item detection, characterized in that: Obtained by using the training method as described in any one of claims 1-5, wherein the image recognition model includes a region proposal network.

7. A method for detecting illegal items, characterized in that: The following steps are involved: P1. Obtain an image of the object to be detected; P2. Based on the acquired image, use the image recognition model as described in claim 6 to identify whether the object to be detected is an illegal item.

8. A device for detecting illegal items, characterized in that: include: An X-ray unit or a millimeter wave scanning unit for acquiring an image of the object to be detected; The detection module is configured with the image recognition model as described in claim 6, and identifies whether the object to be detected is an illegal item based on the acquired image.

9. An autonomous mobile robot that moves autonomously according to a preset program, characterized in that: It includes a device for detecting illegal items as described in claim 8, so that illegal items can be detected during autonomous movement.

10. An autonomous mobile robot as claimed in claim 9, characterized in that: Also included is an alarm system to sound a warning when illegal items are detected.

Images