Visual integration system capable of carrying deep learning model to realize target detection and identification

Abstract

The invention discloses a visual integrated system capable of carrying a deep learning model to realize target detection and recognition, relates to the fields of mechanical engineering, electronic engineering and artificial intelligence, and aims to solve the problems that an existing visual detection system is single in detection target and complex in structure. The peripheral module is used for outputting a model switching signal to the processor; the camera module is used for collecting a target image and inputting the target image into the selected corresponding trained deep learning model in the processor in the main control module; a plurality of trained deep learning models are arranged in the processor, the processor is used for selecting the corresponding trained deep learning model from the trained deep learning models according to the received model switching signal, and the model processes the target image and outputs a control signal to the peripheral module; and the peripheral module is controlled to output an identification result in light, electric signals, images or other forms. The method is used for one-machine multi-type target image recognition.

Description

Technical Field

[0001] The present invention relates to the fields of mechanical engineering, electronic engineering, and artificial intelligence, and belongs to an image recognition system. Background Art

[0002] To ensure the product quality of the production line, it is necessary to judge whether there are products on the assembly line, whether the categories are correct, whether there are defects, etc. At present, this judgment process mostly relies on manual labor. In a multi-production line workshop, inspection personnel need to be equipped on each production line, resulting in high labor costs. Manual inspection is prone to missed inspection and misjudgment due to human reasons such as fatigue and lack of concentration. At the same time, the reaction time of people limits the detection speed of the production line. Existing vision detection systems can only detect whether there are defects in the same type of product, with a single detection target and a complex structure. Summary of the Invention

[0003] The purpose of the present invention is to solve the problems of single detection target and complex structure of existing vision detection systems, and to propose a vision integration system that can carry a deep learning model to achieve target detection and recognition.

[0004] A vision integration system that can carry a deep learning model to achieve target detection and recognition, the system includes a camera module, a main control module, and a peripheral module; the main control module includes a processor;

[0005] The peripheral module is used to output a model switching signal to the processor;

[0006] The camera module is used to collect a target image and input it into the corresponding trained deep learning model selected in the processor in the main control module;

[0007] Multiple trained deep learning models are set in the processor, which are used to select the corresponding trained deep learning model from multiple trained deep learning models according to the received model switching signal. This model processes the target image and outputs a control signal to the peripheral module to control the peripheral module to output the recognition result in the form of light, electrical signal, image or other forms.

[0008] Preferably, the peripheral module includes a mode switching module;

[0009] The mode switching module is used to output a model switching signal to the processor.

[0010] Preferably, the peripheral module further includes a relay module;

[0011] The control signal output by the processor is transmitted to an external computer through the relay module to achieve the output of the recognition result in the form of an electrical signal.

[0012] Preferably, the peripheral module further includes a signal indicator light;

[0013] The control signal output by the processor controls the lighting of the control signal indicator light, realizing the output of the recognition result in the form of light.

[0014] Preferably, the main control module further includes a first serial port module;

[0015] The recognition result output by the processor and the target image are sequentially transmitted to the host computer for display through the first serial port module and the peripheral module.

[0016] Preferably, the peripheral module further includes a Type-C interface;

[0017] The recognition result and the target image output by the first serial port module are transmitted to the host computer for display through the Type-C interface.

[0018] Preferably, the peripheral module is further configured to switch the download mode and the working mode of the processor when the processor is started.

[0019] Preferably, the peripheral module further includes a BOOT button;

[0020] When the processor is started, the switching of the download mode and the working mode of the processor is realized by using the BOOT button.

[0021] Preferably, the peripheral module further includes an Ethernet interface; the main control module further includes a second serial port module;

[0022] The processor communicates with external devices through the second serial port module and the Ethernet interface in sequence.

[0023] Preferably, the system further includes an upper half shell, a lower half shell and a heat sink;

[0024] The peripheral module is arranged inside the lower half shell, the main control module and the camera module are both arranged on the peripheral module, the heat sink is arranged on the processor of the main control module, the camera module extends out of the upper half shell through the through hole on the upper half shell, and the upper half shell is buckled on the lower half shell.

[0025] The beneficial effects of the present invention are:

[0026] By inputting multiple deep learning models into the main control module of the present invention, it is realized that only one set of system can complete the detection of various types of targets on the production line; it can not only classify the products on the production line, but also detect the abnormalities and defects of the products.

[0027] The present invention adopts a mode switching module, and the system detection target is switched by the operator changing the position of the push rod of the mode switching module to the corresponding position;

[0028] The present invention is only composed of three modules: a camera module, a main control module and a mode switching module, with a simple structure and high system integration, and can be adaptively installed on various production lines, occupying a small space on the production line. Description of the Drawings

[0029] Figure 1 It is a schematic diagram of the principle of a vision integration system that can carry a deep learning model to achieve object detection and recognition;

[0030] Figure 2 It is an exploded view of the structure of a vision integration system that can carry a deep learning model to achieve object detection and recognition;

[0031] Figure 3 It is a position relationship diagram of a vision integration system that can carry a deep learning model to achieve object detection and recognition. Detailed Implementation Modes

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0034] Next, the present invention will be further described in conjunction with the accompanying drawings and specific embodiments, but it is not a limitation of the present invention.

[0035] Embodiment 1:

[0036] Combined with Figures 1 to 3 To illustrate this embodiment, a vision integration system that can carry a deep learning model to achieve object detection and recognition, the system includes a camera module 1, a main control module 2 and a peripheral module 3; the main control module 2 includes a processor 2-1;

[0037] The peripheral module 3 is used to output a model switching signal to the processor 2-1;

[0038] The camera module 1 is used to collect a target image and input it into the corresponding trained deep learning model selected in the processor 2-1 in the main control module 2;

[0039] Multiple trained deep learning models are set in the processor 2-1, which are used to select the corresponding trained deep learning model from multiple trained deep learning models according to the received model switching signal. This model processes the target image and outputs a control signal to the peripheral module 3 to control the peripheral module 3 to output the recognition result in the form of light, electrical signal, image or other forms.

[0040] Specifically, only one model is set in the existing processor, and only one type of image can be recognized. However, in the processor of this embodiment, multiple models are preset. The multiple models can respectively predict different recognition results such as whether there is a bottle and whether there is a label on the bottle. Therefore, with one set of systems in this embodiment, various results on the production line can be recognized, and there is no need to equip other recognition instruments, saving the cost of the instruments.

[0041] A method for implementing the selection of the corresponding trained deep learning model is given: The mode switching module 3-1 outputs a model switching signal to the processor 2-1, and the processor 2-1 selects the corresponding trained deep learning model from multiple trained deep learning models according to the received model switching signal.

[0042] The peripheral module 3 also has other functions and implementation methods, which are introduced as follows:

[0043] The peripheral module 3 further includes a relay module 3-2;

[0044] The control signal output by the processor 2-1 is transmitted to the external computer through the relay module 3-2 to output the recognition result in the form of an electrical signal.

[0045] The peripheral module 3 further includes a signal indicator light 3-6;

[0046] The control signal output by the processor 2-1 controls the signal indicator light 3-6 to light up, so as to output the recognition result in the form of light.

[0047] Specifically, if this system is used to recognize whether there is a label on a bottle, when it is recognized that a certain bottle has no label, the processor will output a control signal to be transmitted to the industrial control computer through the relay module, and the industrial control computer controls the external mechanical equipment to act, blowing off or removing the bottle without a label. At the same time, the signal indicator light 3-6 lights up to prompt the staff that there is a bottle without a label.

[0048] The peripheral module 3 is also used to switch the download mode and working mode of the processor 2-1 when the processor 2-1 is started.

[0049] The peripheral module 3 further includes a BOOT button 3-3;

[0050] When the processor 2-1 is started, the download mode and working mode of the processor 2-1 are switched by using the BOOT button 3-3.

[0051] Specifically, the download mode is to download the trained deep learning model into the main control module 2, and the recognition mode is that the trained deep learning model receives the target image and outputs the recognition result.

[0052] The peripheral module 3 further includes an Ethernet interface 3-5; the main control module 2 further includes a second serial port module 2-3;

[0053] The processor 2-1 communicates with external devices through the second serial port module 2-3 and the Ethernet interface 3-5 in sequence.

[0054] Specifically, when the Type-C of the vision integration system is connected to a computer, the Type-C port provides power for the system and can also transmit the current image and detection results to the host computer. At this time, it is possible to check whether the system is running properly and adjust the camera focus.

[0055] Further define the interface for connecting to the host computer:

[0056] The main control module 2 further includes a first serial port module 2-2; the peripheral module 3 further includes a Type-C interface 3-4;

[0057] The processor 2-1 is also used to transmit the output result and the target image to the host computer for display through the first serial port module 2-2 and the Type-C interface 3-4 in sequence.

[0058] Specifically, it is more convenient to transmit the output result and the target image through the Type-C interface 3-4.

[0059] Next, further define the composition of the camera module: The camera module 1 includes a seventh power supply module, a photosensitive chip, and a lens module;

[0060] The seventh power supply module is used to provide voltage for the photosensitive chip;

[0061] The lens module is used to collect the reflected light of the target and focus it on the photosensitive chip;

[0062] The photosensitive chip is used to receive voltage, image the target, and generate a target image.

[0063] Next, a method for building the structure of a vision integration system that can carry a deep learning model to achieve target detection and recognition is given, as Figure 2 and 3 shown:

[0064] The integrated circuit board consists of a main control module 2 and a peripheral module 3. The main control module 2 and the peripheral module 3 are electrically connected through pin headers and sockets. The lower half shell 5 is connected to the peripheral module 3 through screws and copper posts. The peripheral module 3 is connected to the camera module 1 through screws and copper posts. The camera module 1 is connected to the main control module 2 through a flexible cable. The upper half shell 4 is connected to the lower half shell 5 through screws and copper posts. The heat sink 6 is pasted on the processor of the main control module 2. In the peripheral module 3, a power indicator 3-7, a relay module 3-2, a BOOT button 3-3, a signal indicator 3-6, a mode switching module 3-1, a Type-C interface 3-4, and an Ethernet interface 3-5 are carried through circuit traces. At the same time, the peripheral module 3 integrates and carries a first power module. The main control module 2 integrates and carries a second power module, a third power module, a fourth power module, a fifth power module, a sixth power module, a Flash circuit, a processor, a first serial port module, and a second serial port module. The camera module 1 integrates and carries a seventh power module, a photosensitive circuit, and a lens module.

[0065] Working principle:

[0066] When in use, power is supplied to the vision integration system through the Type-C interface 3-4, and the power indicator 3-7 lights up. Press the BOOT button 3-3 to make the vision integration system enter the download mode. Load the deep learning model formed according to the detected target data pictures and videos onto the main control module 2. When the detected targets are multiple different targets, multiple deep learning models can be loaded simultaneously. Select the detection target by toggling the push rod of the mode switching module 3-1. Re-plug the Type-C interface 3-4 to re-power the vision integration system, and the system resumes to the recognition mode. The system performs corresponding target recognition and detection. When the output result of the camera module 1 is inconsistent with the preset result, a control signal is output to control the state change of the relay module 3-2, and at the same time, the signal indicator 3-6 lights up. If the Type-C 3-4 interface is connected to the host computer at this time, while the host computer supplies power to the vision integration system, turn on the host computer. The host computer obtains images from the camera module 1 and displays the recognition results.

[0067] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not deviate from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A vision integration system capable of carrying a deep learning model to achieve object detection and recognition, characterized in that, The system includes a camera module (1), a main control module (2), and a peripheral module (3); the main control module (2) includes a processor (2-1); The peripheral module (3) is configured to output a model switching signal to the processor (2-1); The camera module (1) is configured to collect a target image and input it into a corresponding trained deep learning model selected in the processor (2-1) within the main control module (2); A plurality of trained deep learning models are provided in the processor (2-1), which is configured to select a corresponding trained deep learning model from the plurality of trained deep learning models according to the received model switching signal, process the target image, and output a control signal to the peripheral module (3) to control the peripheral module (3) to output an identification result in the form of light, electrical signal, image, or other forms.

2. The visual integration system capable of carrying a deep learning model to implement target detection and recognition according to claim 1, wherein, The peripheral module (3) includes a mode switching module (3-1); The mode switching module (3-1) is used to output a model switching signal to the processor (2-1).

3. The visual integration system according to claim 1, which can carry a deep learning model to implement target detection and recognition, is characterized in that The peripheral module (3) further includes a relay module (3-2); The control signal output by the processor (2-1) is transmitted to an external computer through the relay module (3-2) to output the identification result in the form of an electrical signal.

4. The visual integration system according to claim 3, which is capable of carrying a deep learning model to achieve target detection and recognition, is characterized in that The peripheral module (3) further includes a signal indicator light (3-6); The control signal output by the processor (2-1) controls the signal indicator light (3-6) to light up to output the identification result in the form of light.

5. The visual integration system capable of carrying a deep learning model to achieve target detection and recognition according to claim 4, characterized in that, The main control module (2) further includes a first serial port module (2-2); The identification result and the target image output by the processor (2-1) are sequentially transmitted to a host computer for display through the first serial port module (2-2) and the peripheral module (3).

6. The visual integration system capable of carrying a deep learning model to implement target detection and recognition according to claim 5, wherein The peripheral module (3) further includes a Type-C interface (3-4); The identification result and the target image output by the first serial port module (2-2) are transmitted to a host computer for display through the Type-C interface (3-4).

7. The visual integration system capable of carrying a deep learning model to implement target detection and recognition according to claim 1 or 6, characterized in that The peripheral module (3) is further configured to switch the download mode and working mode of the processor (2-1) when the processor (2-1) is started.

8. The visual integration system according to claim 7, which is capable of carrying a deep learning model to implement target detection and recognition, is characterized in that The peripheral module (3) further includes a BOOT button (3-3); When the processor (2-1) is started, the download mode and working mode of the processor (2-1) are switched by using the BOOT button (3-3).

9. The visual integration system according to claim 8, which is capable of carrying a deep learning model to implement target detection and recognition, is characterized in that, The peripheral module (3) further includes an Ethernet interface (3-5); the main control module (2) further includes a second serial port module (2-3); The processor (2-1) communicates with external devices sequentially through the second serial port module (2-3) and the Ethernet interface (3-5).

10. The visual integration system capable of carrying a deep learning model to implement target detection and recognition according to claim 9, wherein, The system further includes an upper half housing (4), a lower half housing (5), and a heat sink (6); The peripheral module (3) is arranged inside the lower half shell (5), both the main control module (2) and the camera module (1) are arranged on the peripheral module (3), the heat sink (6) is arranged on the processor (2-1) of the main control module (2), the camera module (1) extends out of the upper half shell (4) through the through hole on the upper half shell (4), and the upper half shell (4) is buckled on the lower half shell (5).

Images