CN211877812U - A glass detection device based on deep learning - Google Patents

Abstract

The utility model discloses a glass detection device based on degree of depth study, including fixed glass transfer unit and collection image part, fixed glass transfer unit includes the conveyer belt and is used for fixed glass's fixing device, fixing device is including two spinal branch props the iron set up two transparent draw-in groove and light filling lamp and setting on the support iron set are in gyro wheel on the conveyer belt, and set up control button and flexible flat board on the gyro wheel, it is in including matching the setting to gather the image part fixed glass transfer unit's industry camera. According to the method, after the glass to be detected is sent to the image acquisition part through the conveyor belt, the distance between the high-definition camera and the glass is adjusted through the position sensor, and the fact that the industrial camera can extract complete, clear and effective glass information is guaranteed; adopt this application device to replace artifical and traditional detection device, can promote glass degree of automation, improve detection efficiency by a wide margin, reduce the cost of labor.

Description

A glass detection device based on deep learning

technical field

The utility model relates to the technical field of glass detection, in particular to a glass detection device based on deep learning.

Background technique

Glass is widely used in building materials and its products due to its beauty, practicability, wind pressure resistance, cold and heat resistance, impact resistance, light transmission and other advantages. Among them, glass, as the core component of the appearance of various buildings, reflects the architectural In order to meet the effect of diversified decoration styles in the implementation of building decoration, it is necessary to carry out defect detection on glass. The existing glass detection devices have low efficiency, and are prone to false detection and missed detection, which are not suitable for popularization and application.

Utility model content

The technical problem to be solved by the present invention is to provide a glass detection device based on deep learning, which is convenient to improve the glass detection efficiency and can reduce the probability of false detection and missed detection.

In order to solve the above-mentioned technical problems, the scheme of the present utility model is:

A glass detection device based on deep learning, comprising a fixed glass conveying part and an image capturing part, the fixed glass conveying part comprises a conveyor belt and a fixing device for fixing the glass, the fixing device comprises two erected supporting iron bars, and a The transparent card slots and the fill light on the two supporting iron rods, the rollers arranged on the conveyor belt, the control buttons and the telescopic flat plate arranged on the rollers, and the image capturing component includes matching and arranged on the The industrial camera for fixing the glass conveying part, the automatic photosensitive fill light arranged outside the industrial camera, and the horizontal slide rail connected above the industrial camera, and the position set on the horizontal slide rail A sensor, the industrial camera is communicatively connected to an external deep learning detection module, and the position sensor is communicatively connected to the industrial camera.

The transparent card slot is fixed on the support iron rod by a knob, the transparent card slot is made of transparent material, and the edges of the transparent card slot are provided with clipping teeth at intervals.

It also includes an automatic device for adjusting the width of the transparent card slots on both sides. The automatic device is composed of the control button, a telescopic flat plate and a roller, and the roller is provided with a buckle.

The control button is connected with the telescopic flat plate and the roller control through the built-in weak electric traction.

The deep learning detection module includes a deep learning chip, a pre-trained AlexNet network and a set UI interface.

The pre-trained AlexNet network is a pre-trained Alex-Net network model loaded using TensorFlow.

The deep learning chip is any one of a GPU chip, an FPGA chip or an ASIC chip.

Compared with the prior art, the beneficial effects of the present utility model are:

In this application, after the glass to be tested is sent to the image acquisition part through the conveyor belt, the distance between the high-definition camera and the glass is adjusted by the position sensor to ensure that the industrial camera can extract complete, clear and effective glass information, and the physical surface of the glass to be tested is collected by the high-definition camera. The information is sent to the pre-trained AlexNet network to detect glass physical surface defects. After pre-training, the AlexNet network can automatically extract the physical surface defect information of the glass to be tested (including scratches, spots, bubbles, clarity, defects, etc.) and The results are displayed on the UI interface. When the test passes, the quality of the glass to be tested meets the standard, and the glass quality inspection is completed. Using the device of the present application to replace manual and traditional detection devices can improve the degree of glass automation, greatly improve the detection efficiency, and reduce labor costs.

Description of drawings

Fig. 1 is the structural representation of the utility model;

Fig. 2 is the side view of the utility model conveyor belt;

FIG. 3 is a top view of the glass card slot of the present invention.

Detailed ways

The specific embodiments of the present utility model will be further described below with reference to the accompanying drawings. It should be noted here that the description of these embodiments is used to help the understanding of the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as there is no conflict with each other.

As shown in Figures 1-3, a deep learning-based glass detection device includes a fixed glass conveying part and an image capturing part, the fixed glass conveying part includes a conveyor belt 9 and a fixing device for fixing the glass, and the fixing device includes The two erected supporting iron bars 4, the transparent card slots 5 and the fill light 6 arranged on the two supporting iron bars 4, and the rollers 10 arranged on the conveyor belt 9, and the rollers 10 arranged on the The control button 7 and the telescopic flat plate 8, the image capturing component includes an industrial camera 1 matched with the fixed glass conveying component, and an automatic photosensitive fill light 2 arranged outside the industrial camera 1, and connected to the The horizontal slide rail 11 above the industrial camera 1, and the position sensor 3 arranged on the horizontal slide rail 11, the industrial camera 1 is communicatively connected to an external deep learning detection module, and the position sensor 3 is connected to the The communication connection of the industrial camera 1 is described.

The transparent card slot 5 is fixed on the supporting iron rod 4 by a knob, and the upper and lower widths of the transparent card slot 5 can also be adjusted by a knob. The height of the irons needs to be consistent.

The transparent card slot 5 is made of transparent material, and the edges of the transparent card slot 5 are provided with locking teeth at intervals. This setting can facilitate the maximum area to ensure that the surface features of the glass to be tested are collected.

It also includes an automatic device for adjusting the width of the transparent card slots 5 on both sides. The automatic device is composed of the control button 7, the telescopic flat plate 8 and the roller 10, and the roller 10 is provided with a buckle. This setting is mainly to meet the needs of various glass widths to be tested and to enhance the practicability of the glass detection device.

The control button 7 is controlled and connected with the telescopic flat plate 8 and the roller 10 through the built-in weak current traction. This setting is mainly to facilitate control adjustment and fixed positioning. When in use, when the control button 7 is pressed for the first time, the telescopic plate 8 can be adjusted to extend and shorten, and the roller 10 provided with the buckle can move left and right to speed up the adjustment of the left and right support irons 4. When the control button 7 is pressed again, the telescopic plate can be adjusted. 8 and the roller 10 will be fixed, so as to adjust the left and right width of the transparent card slot 5 to meet the needs of different types of glass to be tested.

The deep learning detection module includes a deep learning chip, a pre-trained AlexNet network and a set UI interface. With the help of the pre-trained AlexNet network, the chip GPU can better automatically identify defects. Since the image acquisition device is connected to the designed UI interface through the deep learning chip, the UI interface can display the input picture and mark the defect position with a red frame.

The pre-trained AlexNet network is a pre-trained Alex-Net network model loaded using TensorFlow. This setting is mainly used for data analysis and processing through the Alex-Net network, including multi-layer convolution layers, pooling layers, fully connected layers, and information feedback through the back-propagation algorithm, learning to obtain convolution parameters (that is, with identification defects Ability).

The deep learning chip is any one of a GPU chip, an FPGA chip or an ASIC chip. In the application process, GPU, FPGA, and ASIC are all existing mainstream deep learning chips, which can be directly purchased in the market according to requirements.

In this application, after the glass to be tested is sent to the image acquisition part through the conveyor belt 9, the distance between the high-definition camera and the glass is adjusted by the position sensor 3 to ensure that the industrial camera 1 can extract complete, clear and effective glass information, and the high-definition camera collects the glass to be tested. The physical surface information is sent to the pre-trained AlexNet network to detect glass physical surface defects. After pre-training, the AlexNet network can automatically extract the physical surface defect information of the glass to be tested (including scratches, spots, bubbles, clarity, defects, etc. ) and display the results through the UI interface. When the test passes, the quality of the glass to be tested reaches the standard, and the glass quality inspection is completed. Using the device of the present application to replace manual and traditional detection devices can improve the degree of glass automation, greatly improve the detection efficiency, and reduce labor costs.

In this application, the number of supporting iron bars 4 should be appropriately set according to the width, weight and length of the glass to be tested. at the same time. In order to facilitate automatic ejection, a retraction spring is provided at the root of the support iron rod 4 for fixing the glass conveyor belt part. When the support iron rod 4 starts from the starting point, it will automatically pop up vertically on the conveyor belt 9, and when the transmission reaches the end point, the support iron rod will shrink and lie flat on the conveyor belt. to complete the operation of the entire conveyor belt.

The high-definition camera 1 is moved up and down through the horizontal slide rail 11 to ensure that the industrial camera 1 can clearly capture glass images of different sizes. The horizontal slide rail 11 is equipped with a position sensor, and the lower side of the horizontal slide rail is mainly used to fix the industrial camera 1 to complete image acquisition and fix the position sensor to complete the adjustment of the distance between the industrial camera and the glass to be measured.

The outside of the lens of the industrial camera 1 is provided with an automatic photosensitive fill light 2, which is convenient to automatically adjust the brightness of the light; the conveyor belt 9 is under the buckle roller, and the width of the conveyor belt must meet the maximum width of the glass to be tested on the market.

In this application, the surface defects of the glass to be tested include: scratch-type defects, spot-type defects (surface spots, built-in bubbles), defects (edge and corner defects), point defects (black spots, defects, stains), clarity (turbidity, Vague);

In the glass physical surface inspection of the present application, the image acquisition component collects a clear image to be tested, and then performs image processing based on deep learning, and marks the defect position according to the difference between the defect position of the image and the normal image.

The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, without departing from the principle and spirit of the present invention, various changes, modifications, substitutions and alterations can be made to these embodiments, which still fall within the protection scope of the present invention.

Claims (7)

1. The utility model provides a glass detection device based on degree of deep learning which characterized in that: the device comprises a fixed glass conveying part and an image collecting part, wherein the fixed glass conveying part comprises a conveying belt (9) and a fixing device used for fixing glass, the fixing device comprises two vertical supporting iron rods (4), a transparent clamping groove (5) and a light supplement lamp (6) which are arranged on the two supporting iron rods (4), a roller (10) arranged on the conveying belt (9), a control button (7) and a telescopic flat plate (8) which are arranged on the roller (10), the image collecting part comprises an industrial camera (1) which is arranged on the fixed glass conveying part in a matching manner, an automatic photosensitive light supplement lamp (2) arranged on the outer side of the industrial camera (1), a horizontal sliding rail (11) connected above the industrial camera (1) and a position sensor (3) arranged on the horizontal sliding rail (11), the industrial camera (1) is in communication connection with an external deep learning detection module, and the position sensor (3) is in communication connection with the industrial camera (1).

2. The deep learning based glass inspection device of claim 1, wherein: the transparent clamping groove (5) is fixed on the supporting iron rod (4) through a knob, the transparent clamping groove (5) is made of transparent materials, and clamping teeth are arranged at the edge of the transparent clamping groove (5) at intervals.

3. The deep learning based glass inspection device of any of claims 1-2, wherein: the automatic device is characterized by further comprising an automatic device used for adjusting the width of the transparent clamping groove (5) at two sides, the automatic device is composed of a control button (7), a telescopic flat plate (8) and a roller (10), and the roller (10) is provided with a buckle.

4. The deep learning based glass inspection device of claim 1, wherein: the control button (7) is in control connection with the telescopic flat plate (8) and the roller (10) through built-in weak current traction.

5. The deep learning based glass inspection device of claim 1, wherein: the deep learning detection module comprises a deep learning chip, a pre-trained AlexNet network and a set UI interface.

6. The deep learning based glass inspection device of claim 5, wherein: the pre-trained AlexNet network is an Alex-Net network model loaded and pre-trained by using TensorFlow.

7. The deep learning based glass inspection device of claim 5, wherein: the deep learning chip is any one of a GPU chip, an FPGA chip or an ASIC chip.

Images