TWI734297B - Multi-task object recognition system sharing multi-range features - Google Patents

Abstract

The invention discloses a multi-task object recognition system sharing multi-range features. First, a multi-range generation module generates multi- range information of an image of an object. Then, a multi-channel image merge module samples multiple images in different regions from the object according to the multi-range information of the image of the object to combine the multiple images into one multi-channel image. Thereafter, a shared feature extraction module with a shared network layer extracts shared features from the multi-channel image using a neural network. Final, one or more different specific task feature extraction modules of a task-specific feature extraction module group extract one or more different specific task features from the sharing feature using the neural network, and one or more different specific task models output a recognition result of the image of the object according to the one or more different specific task features.

Description

Multi-task object recognition system sharing multi-range features

The present invention relates to an object identification technology, in particular to a multi-task object identification system that shares multiple range features.

In view of the vigorous development of deep learning technology, most of today’s image recognition tasks can be achieved through the training of neural-like models, but practical application scenarios often require the integration of multiple recognition tasks. If each recognition task is based on its own model Responsible, the computational complexity will increase with the number of identification tasks. Take face recognition as an example. The application site is often under strict authority control, so it needs to interact and match both the face recognition task and the fake face recognition task. However, if the face recognition task and the fake face recognition task use two independent Model, the computational complexity will also double.

Recently, in order to solve the problem of increased efficiency due to increased tasks, a single model is responsible for the integration of different tasks. However, because each image recognition task has different requirements for the area range of the object, the forced merging will cause the recognition of each task to be correct. The rate drops. For example, although both face recognition and fake face recognition use face images as input, face recognition pays attention to the facial features of the face and requires a small range of images, while fake face recognition considers the background ring around the face. Due to the environment and the required image range is relatively large, it is difficult to determine the input range of the image due to both face recognition and fake face recognition.

Therefore, how to provide a novel or innovative object (image) recognition technology to solve the integration problem of multiple recognition tasks has become a major research topic for those skilled in the art.

The present invention provides a novel or innovative multi-task object recognition system that shares multi-range features. For example, it can solve the problem that multiple independent models are not easy to integrate when the same object is used in multi-task applications, or provide a shared network layer for effective Reduce repetitive or unnecessary network layers, or allow multiple recognition tasks to share features extracted from multi-range images to improve recognition accuracy.

The multi-task object recognition system that shares multi-range features in the present invention includes: a multi-range generation module that generates or provides multi-range information of the image of the object; a multi-channel image merging module that is based on the multi-range generation module The generated or provided multi-range information of the image of the object samples multiple images of different areas from the object to merge the multiple images into a multi-channel image; a shared feature extraction module with a shared network layer, is A neural network is used to extract shared features from the multi-channel images merged by the multi-channel image merging module; and a task-specific feature extraction module group has one or more different task-specific feature extraction modules to Use neural networks to extract one or more different specific task features from the shared features extracted by the shared feature extraction module with the shared network layer, so as to extract one or more specific tasks of the module group from the specific task feature The model outputs the recognition result of the image of the object according to one or more different specific task features.

In order to make the above-mentioned features and advantages of the present invention more comprehensible, embodiments are specifically described below in conjunction with the accompanying drawings. In the following description, the additional features and advantages of the present invention will be partially explained, and these features and advantages will be partly known from the description, or can be learned by practicing the present invention. The features and advantages of the present invention are realized and achieved by means of the elements and combinations specifically pointed out in the scope of the patent application. It should be understood that both the foregoing general description and the following detailed description are only illustrative and explanatory, and are not intended to limit the scope of the present invention.

1‧‧‧Multi-task object recognition system with shared multi-range features

10‧‧‧Image capture module

20‧‧‧Object Detection Module

30‧‧‧Multi-range generation module

40‧‧‧Multi-channel image merging module

50‧‧‧Shared feature extraction module

51‧‧‧Shared network layer

52‧‧‧Shared feature map

60‧‧‧Special task feature extraction module group

61‧‧‧Special task feature extraction module

62‧‧‧Specific mission model

70‧‧‧Cross-task Application Service Module

A‧‧‧Video or video screen

B‧‧‧Location Information

C‧‧‧Multi-channel image

C1 to Cn‧‧‧Image

F‧‧‧Characteristic map

L0‧‧‧Level 0 shared network layer

L1‧‧‧The first shared network layer

Lk‧‧‧k-th shared network layer

m, n‧‧‧positive integer

Figure 1 is a schematic diagram of the architecture of the multi-task object recognition system sharing multiple range features in the present invention;

Figure 2 is a schematic diagram of an embodiment of a multi-task object recognition system that shares multi-range features in the present invention;

Figure 3 is a schematic diagram of an embodiment of the input of the image capture module, the object detection module, the multi-range generation module, and the multi-channel image merging module in Figures 1 to 2 of the present invention; and

Figure 4 is a schematic diagram of an embodiment of the shared network layer of the shared feature extraction module in Figure 2 of the present invention.

The following describes the implementation of the present invention with specific specific embodiments. Those familiar with this technology can understand the other advantages and effects of the present invention from the content disclosed in this specification. It can also be implemented or applied by other different specific equivalent embodiments.

Fig. 1 is a schematic diagram of the architecture of the multi-task object recognition system 1 sharing multiple-range features in the present invention, and Fig. 2 is a schematic diagram of an embodiment of the multi-task object recognition system 1 sharing multiple-range features in the present invention.

As shown in Fig. 1 and Fig. 2, the multi-task object recognition system 1 sharing multi-range features may include an image capturing module 10, an object detection module 20, and a multi-range generation module that communicate with each other. 30. A multi-channel image merging module 40, a shared feature extraction module 50, a specific task feature extraction module group 60, and a cross-task application service module 70. The shared feature extraction module 50 may have a shared network layer 51 (such as the 0th shared network layer L0 to the kth shared network layer Lk, and the task-specific feature extraction module group 60 may have one or more different Specific task feature extraction module 61.

The image capture module 10 can capture images or image frames, and the object detection module 20 can detect the position information of objects from the images or image frames captured by the image capture module 10, and a multi-range generation module 30 can use the position information of the object detected by the object detection module 20 to generate or provide multi-range information of the image of the object. The multi-channel image merging module 40 can sample multiple images of different regions and ranges from the object according to the multi-range information of the image of the object generated or provided by the multi-range generation module 30 to merge the multiple images into one multi-channel image. The shared feature extraction module 50 can extract shared features from the multi-channel image merged by the multi-channel image merging module 40 by using a neural network. One or more different task-specific feature extraction modules 61 of the task-specific feature extraction module group 60 can be extracted from the shared features extracted by the shared feature extraction module 50 with the shared network layer 51 by using a neural network. One or more different specific task features, so that one or more specific task models 62 of the specific task feature extraction module group 60 are based on one Or multiple different specific task feature output object image recognition results.

For example, two different task-specific models 62 are a face recognition model and a fake face recognition model respectively, which are used as the application of the anti-counterfeiting face recognition access control system, and the cross-task application service module 70 can use the specific task feature extraction module group 60 One or more specific task features extracted by one or more different specific task feature extraction modules 61 provide, execute, or achieve cross-task application services. That is, each task in the specific task feature extraction module group 60 corresponds to a specific task model 62, and the specific task model 62 is composed of a neural network and extracts required features from the shared feature map 52 to determine the input What are the characteristics of the image of, in terms of face, suppose there are two specific task models 62 which are face recognition model and gender recognition model respectively, then these two specific task models 62 extract the required features from the shared feature map 52 After that, the recognition result of the image of the object can be output. For example, after the image of the President of the United States is processed by the image capture module 10 to the specific task feature extraction module group 60, the two specific task models 62 will output "Trump", Two pieces of information such as "male" are given to the cross-task application service module 70.

The image capturing module 10 can be a hardware camera, camera, monitor, or sensor, etc., and the image captured by the image capturing module 10 can be a two-dimensional image or a three-dimensional image, such as a depth image, an infrared image And so on, and the channel of the two-dimensional image can be a single channel or multiple channels.

The object detection module 20 can be a hardware object detector or a software object detection program, etc., and the object detection module 20 can use different object detectors according to different applications. For example, face analysis or recognition applications can use object detectors or object detection programs related to MTCNN (Multi-task cascaded convolutional networks), and vehicle detection or recognition applications can Use the object detector or object detection program related to YOLO (You Only Look Once).

The multi-range generating module 30 may be a software multi-range generating program, etc., and the multi-channel image merging module 40 may be a software multi-channel image merging program, etc. The multi-range expansion method and quantity of the multi-range generation module 30 or the multi-channel image merging module 40 can vary according to different applications. If the details and texture characteristics of the object are emphasized, the multi-range generation module 30 or multi-channel The image merging module 40 can be based on multiple reduced-range images; on the contrary, if it focuses on the appendages or background around the object, the multi-range generation module 30 or the multi-channel image merging module 40 uses multiple expanded-range images. The image is the main one (that is, the image with more expanded range is added). In addition, if the focus is on the size or performance of the model of the shared network layer 51, the multi-range generation module 30 or the multi-channel image merging module 40 can overlap the number of images of different areas (that is, reduce the number of overlapped images). The number of images); on the contrary, if the focus is on improving the versatility of the shared network layer 51, the multi-range generation module 30 or the multi-channel image merging module 40 can superimpose the number of images of more different ranges (In other words, increase the number of superimposed images) to facilitate capturing more features of the image.

The shared feature extraction module 50 can be a software shared feature extraction program, etc., and the type and number of the shared network layer 51 (neural network layer) of the shared feature extraction module 50 can be adjusted according to different applications. For example, in some applications, the shared network layer 51 of the shared feature extraction module 50 may use three-dimensional (3D) convolution to achieve better results, or when multiple task-specific feature extraction modules 61 are quite similar, the shared feature The number of layers of the shared network layer 51 of the extraction module 50 can be increased.

The specific task feature extraction module 61 of the specific task feature extraction module group 60 can be a software specific task feature extraction program, etc., and the cross-task application service module 70 can be a software cross-task application service program, etc. The number of network layers of each specific task feature extraction module 61 does not need to be the same, and there is no limit to the number, and each specific task feature extraction module 61 can be based on the characteristics of the task Use different types and quantities of network layers.

The aforementioned neural network can be, for example, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Deep Neural Network (DNN), Long Short-term Memory (LSTM) ) Neural network, etc., but not limited to this.

For example, the multi-range generation module 30 can capture multiple reduced or expanded image ranges, and the multi-channel image merging module 40 can superimpose multiple images based on the multiple images captured by the multi-range generation module 30. The dimensions of the channel are integrated into a single input, so that the features included in a single input can span multiple recognition tasks. Taking face analysis or recognition as an example, the expanded face image can cover hair and ears, which is helpful for gender judgment and face recognition applications; moreover, the reduced and expanded face images are helpful to capture people. The texture and background characteristics of the face image help the application of living body (such as human body) recognition.

The shared feature extraction module 50 may have a shared network layer 51, which can effectively reduce repetitive or unnecessary network layers, so as to reduce the size of the model and increase the execution speed. In application, the specific task model 62 of the specific task feature extraction module group 60 can also be exchanged according to different needs to improve flexibility. For example, take an example to illustrate the concept of swap. Suppose a product with face recognition, gender recognition and age recognition is designed, but in a certain application scenario, the customer does not need age information, and the specific task of age recognition can be removed The model 62 retains the task-specific model 62 for both face recognition and gender recognition to reduce the overall model size and computational complexity. Therefore, the present invention allows the shared network layer 51 to be shared between multiple recognition tasks, reduces the overall computational complexity, and allows multiple recognition tasks to share features extracted from multiple ranges of images, and improves the recognition accuracy. .

The invention can provide multi-range image overlay, cross-task sharing features and swappable Technology such as feature extraction group, using these technologies can make the application of the same object but multiple identification tasks easy to integrate into a single model, which can effectively reduce the size of the model and the prediction time, and also increase the flexibility of the application. That is, the present invention can utilize the technology of sharing features in multiple ranges, and the application of the same object but multiple identification tasks can be integrated into a single model through the shared network layer 51 of the shared feature extraction module 50, so as to effectively reduce the size of the model. And forecast time.

The invention can solve the problems of large model and low efficiency. That is, under multiple independent models, since the range of images required for each recognition task may be slightly different, the object to be recognized is the same, but multiple images with different ranges (such as n images) need to be captured. (Or pictures) to input into multiple (such as n) independent models to reduce the overall recognition speed. Therefore, the present invention uses the technology of multi-range (multi-channel) overlay and shared network layer 51 to integrate the features required by the model into a single model, so that the size of the model can be reduced.

The shared feature extraction module 50 is designed to reduce the amount of calculation and increase the speed of prediction, and the concept is that sharing can reduce the amount of calculation, thereby increasing the speed of prediction. For example, suppose that the calculation amount of independent model X is 50 and the calculation amount of independent model Y is 50. If the prediction is made separately, the total calculation amount of independent model X and independent model Y is 50+50=100, but If the concept of sharing is introduced, the part that can be shared by independent model X and independent model Y is split and called Z, and the calculation amount of this shared part Z is 20, then the remaining independent model X and independent model Y cannot be shared The parts of are called X'and Y'respectively. After sharing, the total calculation amount will become Z+X'+Y'=20+(50-20)+(50-20)=80, compared to the original calculation A quantity of 100 saves 20 calculations.

The invention can have a high degree of application flexibility. That is, the present invention integrates the features required by the model into a single model, and still retains the flexibility of an independent model. Different task-specific feature extraction modules 61 can be exchanged according to different applications, and different task groups can be customized according to different applications. combine.

Figure 3 is a schematic diagram of an embodiment of the input of the image capture module 10, the object detection module 20, the multi-range generation module 30, and the multi-channel image merging module 40 in Figures 1 to 2 of the present invention. .

As shown in Fig. 3 and Figs. 1 to 2 above, the key of the present invention is to generate an image capture module 10, an object detection module 20, a multi-range generation module 30, and a multi-channel image merging module 40 It also provides the shared network layer 51 of the shared feature extraction module 50 (such as the 0th shared network layer L0 to the kth shared network layer Lk), and at least two of the specific task feature extraction module group 60 The specific task model 62 and the cross-task application service module 70 constitute the entire architecture of the system. It should be noted that the aforementioned layer 0 shared network layer L0 means that the shared network layer is counted from 0, but if the shared network layer is counted from 1, the shared network layer L0 of the 0th layer should be changed to the first The layer shares the network layer, and so on.

When generating input from the image capture module 10, the object detection module 20, the multi-range generation module 30, and the multi-channel image merging module 40, the present invention may include the following procedures P11 to P14, and the human face is used as Examples are explained.

Procedure P11: the image or image frame A is captured by the image capturing module 10. For example, the image capturing module 10 (such as a camera or a sensor, etc.) can capture images or image frames A, such as RGB images (pictures) or image frames, where RGB represents the three primary colors of red/green/blue.

Procedure P12: The object detection module 20 detects the position information B of the object from the image or image frame A captured by the image capturing module 10. For example, the object detection module 20 can use an object detection algorithm (such as a face detection algorithm) to detect an object (such as a human face) from the image captured by the image capture module 10 or the image frame A The location information B (such as bounding box coordinates).

Procedure P13: The multi-range generation module 30 uses the position information B of the object detected by the object detection module 20 to generate or provide the multi-range information of the image of the object. For example, the multi-range generation module 30 can expand or reduce the boundary of the object (such as a human face) from the center or other positions of the object (such as a human face) detected by the object detection module 20 according to different applications, and determine the object (Such as a human face) the extent of expansion or reduction and the number of images of the object (such as a human face) to be captured. Subsequently, the multi-range generation module 30 can scale the multiple images C1-Cn of the object (such as a human face) to a fixed length and width after capturing multiple images C1-Cn of the object (such as a human face). Here we take the image of an object (such as a human face) with a fixed length and width of 224x224 (unlimited units) as an example, and assume that the images (such as an RGB image) of objects with different ranges (such as a human face) are multiple images C1- Cn (such as n images), where n is a positive integer.

Procedure P14: The multi-channel image merging module 40 samples multiple images C1-Cn (such as n images) of different regions from the object according to the multi-range information of the image of the object generated or provided by the multi-range generation module 30 ) To merge multiple images C1-Cn into a multi-channel image C. For example, the multi-channel image merging module 40 can sample multiple images C1-Cn (such as n Images) to merge multiple images C1-Cn into one multi-channel image C in the channel dimension. Assuming that the dimension of each image (picture) is 224x224x3, after merging n images (pictures), the dimension of n images is 224x224x3n.

Figure 4 is a schematic diagram of an embodiment of the shared network layer 51 of the shared feature extraction module 50 in Figure 2 of the present invention. As shown in Figures 3 and 4, after the input of the image capture module 10, the object detection module 20, the multi-range generation module 30, and the multi-channel image merging module 40 are generated in Figure 3, Then it will sequentially enter the shared network layer 51 of the shared feature extraction module 50 in Figure 4 The 0th layer shared network layer L0, the first layer shared network layer L1 to the kth layer shared network layer Lk.

As shown in FIG. 4, the shared feature extraction module 50 with the shared network layer 51 can extract shared features from the multi-channel image C merged by the multi-channel image merging module 40 by using a neural network. For example, the shared network layer 51 of the shared feature extraction module 50 can use the following procedures P21 to P22 to utilize the neural network (such as convolutional neural network CNN) merged from the multi-channel image merging module 40 The shared features are extracted from the channel image C. At the same time, the present invention takes the 0th layer shared network layer L0 and two-dimensional (2D) convolution as an example in the following, and the first layer shared network layer L1 to the kth layer shared network layer Lk can be deduced by analogy, and shared The feature extraction module 50 can also select two-dimensional (2D) convolution or three-dimensional (3D) convolution according to different applications.

Procedure P21: The shared network layer 51 of the shared feature extraction module 50 (such as the 0th shared network layer L0) can use (if used in sequence) cores of different length and width (such as convolution cores) for multi-channel images The multi-channel image C merged by the merging module 40 is convolved. It should be noted that when the input channel of shared network layer 51 (such as layer 0 shared network layer L0) is 3n, the two-dimensional (2D) of shared network layer 51 (such as layer 0 shared network layer L0) The depth of the core of the convolution is 3n.

Procedure P22: Assuming that there are a total of multiple (such as m) cores with different lengths and widths, the shared network layer 51 of the shared feature extraction module 50 (such as the shared network layer L0 of the 0th layer) performs the multi-channel image C After convolution, a feature map F with a depth of m can be generated, which can be used as or provided to the next shared network layer (such as the first shared network layer L1), where m is a positive integer.

After the shared feature extraction module 50 generates the feature map F of the shared network layer 51 (such as the 0th shared network layer L0), it can generate a deeper shared network layer (such as the 1st shared network layer) according to different situations. L1 to the k-th shared network layer Lk). For example, one or more specific task models 62 of the specific task feature extraction module group 60 shown in Figure 2 all use ResNet or ResNet system Column, such as ResNet18, ResNet34 or ResNet50, the shared feature extraction module 50 can combine the subsequent batch normalization layer (Batch Normalization Layer), ReLu activation layer (ReLu Activation Function Layer), maximum pooling layer (Max Pooling Layer), etc. Listed in the shared network layer 51 (for example, the 0th shared network layer L0 to the kth shared network layer Lk). Therefore, due to the shared network layer 51 of the shared feature extraction module 50 (such as the 0th layer shared network layer L0 to the kth layer shared network layer Lk), multiple independent models can pass through the shared network layer 51 Integrate into one model to reduce the overall size, that is, the size of a single model integrated through sharing will be smaller than the total size of each independent model.

The shared feature extraction module 50 can generate a shared feature map 52 after calculating all the shared network layers 51 (such as the 0th layer shared network layer L0 to the kth layer shared network layer Lk), so as to convert the shared feature map 52 is input to each specific task model 62 of the specific task feature extraction module group 60 shown in FIG. 2.

For example, due to the reduction and expansion of the multi-range generation module 30 shown in Figure 3, the subtle facial texture features, head surrounding features, and features of objects (such as face/head/background) can be effectively included. The background features, therefore, the face recognition model, the fake face recognition model, the gender recognition model, and the age recognition model are all specific task models 62 suitable for sharing in the specific task feature extraction module group 60 shown in FIG. 2.

As shown in Figures 1 to 2, different task-specific feature extraction modules 61 in the task-specific feature extraction module group 60 output (eg, output in parallel) the results of different task-specific models 62 to the cross-task application service After the module 70, the cross-task application service module 70 can be used with different specific task models 62 for different applications. For example, the cross-task application service module 70 can use the first specific task model 62 (such as a fake face recognition model) to filter out fake human faces first. Then use the recognition results of the second specific task model 62 (such as the face recognition model), the third specific task model 62 (such as the gender recognition model), and the fourth specific task model 62 (such as the age recognition model) to avoid The system was stolen by someone who wanted to.

In summary, the multi-task object recognition system sharing multiple-range features in the present invention can at least have the following features, advantages, or technical effects.

1. The present invention can solve the problem that multiple independent models are not easy to integrate when the same object is used in multi-task applications.

2. The shared feature extraction module of the present invention can have a shared network layer, which can effectively reduce repetitive or unnecessary network layers, so as to reduce the size of the model and increase the execution speed.

3. The present invention allows the shared network layer to be shared among multiple identification tasks, reduces the overall computational complexity, and allows multiple identification tasks to share features extracted from multi-range images, thereby improving the accuracy of identification.

4. The present invention can provide technologies such as multi-range image overlay, cross-task sharing features, and interchangeable feature extraction groups. Using these technologies, applications of the same object but multiple identification tasks can be easily integrated into a single model, which is effective Reducing the size of the model and the prediction time can also increase the flexibility of the application.

5. The present invention can utilize the technology of multi-range sharing of features to integrate the same object but multiple recognition task applications into a single model through the shared network layer of the shared feature extraction module, so as to effectively reduce the size of the model and the prediction time.

6. The present invention uses multi-range (multi-channel) overlay and shared network layer technology to integrate the features required by the model into a single model, so that the size of the model can be reduced, so as to reduce the amount of calculation and the shared feature extraction module prediction. Improve the speed of forecasting.

7. The present invention integrates the features required by the model into a single model, and still retains the flexibility of an independent model. Different specific task feature extraction modules can be exchanged according to different applications, or different task combinations can be customized according to different applications.

8. The application scope of the present invention is quite wide, and it can be used for various image object recognition or monitoring tasks, such as face recognition, living body recognition, gender and age recognition, human figure recognition, license plate recognition, vehicle recognition, image monitoring, smart retail, etc. At the same time, the products that the present invention may be applied to are, for example, facial cleaning products, smart access control products, visitor analysis products, electronic fence products, and the like.

The above embodiments are only illustrative of the principles, features and effects of the present invention, and are not intended to limit the scope of implementation of the present invention. Anyone familiar with the art can comment on the above without departing from the spirit and scope of the present invention. Modifications and changes to the implementation form. Any equivalent changes and modifications made using the content disclosed in the present invention should still be covered by the scope of the patent application. Therefore, the protection scope of the present invention should be as listed in the scope of the patent application.

1‧‧‧Multi-task object recognition system with shared multi-range features

10‧‧‧Image capture module

20‧‧‧Object Detection Module

30‧‧‧Multi-range generation module

40‧‧‧Multi-channel image merging module

50‧‧‧Shared feature extraction module

60‧‧‧Special task feature extraction module group

61‧‧‧Special task feature extraction module

70‧‧‧Cross-task Application Service Module

Claims (9)

A multi-task object recognition system that shares multi-range features includes: a multi-range generation module that generates or provides multi-range information of the object's image; a multi-channel image merging module based on the multi-range generation module The generated or provided multi-range information of the object’s image is to sample multiple images of different areas from the object to merge the multiple images into a multi-channel image; a shared feature extraction model with a shared network layer The group is a multi-channel image merged from multiple images of different regions and ranges sampled by the multi-channel image merging module based on the multi-range information of the object's image of the multi-range generation module by using a neural network The shared features are extracted from the shared feature extraction module, where the shared network layer of the shared feature extraction module uses cores of different length and width. The multi-channel image formed by merging multiple images of different regions sampled by the information is convolved to generate the feature map of the shared network layer; and a specific task feature extraction module group, which has one or more different The specific task feature extraction module of the specific task feature extraction module group from the shared feature extraction module having the shared network layer for the multi-channel image merging module generates the object of the module according to the multi-range One or more different specific task features are extracted from the shared features extracted from the multi-channel image by combining multiple images from different regions and ranges sampled by the multi-range information of the image, so that the specific task feature extraction model One or more specific task models of the group output the recognition result of the image of the object according to the one or more different specific task characteristics. The system described in item 1 of the scope of patent application uses the technology of multi-range sharing of features to integrate applications of the same object but multiple identification tasks into a single model through the shared network layer of the shared feature extraction module . For example, the system described in item 1 of the scope of patent application further includes an image capture module and an object detection module, wherein the image capture module is used to capture images or image frames, and the object detection module The detection module detects the position information of the object from the image or image screen captured by the image capture module, so that the multi-range generation module uses the object detected by the object detection module The location information of produces the multi-range information of the image of the object. For example, the system described in item 1 of the scope of patent application, wherein the multi-range generation module or the multi-range expansion method of the multi-channel image merging module varies according to different applications. If the focus is on the details of the object and Texture feature, the multi-range generation module or the multi-channel image merging module mainly focuses on multiple reduced-range images, and if it focuses on the appendages or background around the object, the multi-range generation module or the The multi-channel image merging module is based on multiple images with an expanded range. For example, the system described in item 1 of the scope of patent application, in which, if the size or performance of the model of the shared network layer is emphasized, the multi-range generation module or the multi-channel image merging module overlaps less different ranges If the versatility of the shared network layer is emphasized, the multi-range generation module or the multi-channel image merging module overlaps the number of images of more different ranges. For the system described in item 1 of the scope of patent application, the multi-range generation module captures multiple reduced or expanded image ranges, and the multi-channel image merging module is based on The multiple images captured by the multi-range generation module are superimposed on the dimensions of multiple channels to be integrated into a single input, so that the features included in the single input span multiple recognition tasks. For example, in the system described in item 1 of the scope of patent application, the shared feature extraction module further includes a batch normalization layer, a ReLu excitation layer, or a maximum pooling layer in the shared network layer. The system described in item 1 of the scope of patent application, wherein the shared feature extraction module generates a shared feature map after calculating all the shared network layers, so as to input the shared feature map to the specific task feature extraction module Each specific task model of the group. For example, the system described in item 1 of the scope of patent application further includes a cross-task application service module that uses the one or more different specific task feature extraction modules of the specific task feature extraction module group to extract the One or more specific task features provide or execute cross-task application services.

Images