Dual-layer printed circuit board component recognition method using dual-channel infrared cloud images

Abstract

The present application relates to a double-layer printed circuit board component recognition method using a double-channel infrared cloud image, comprising the following steps: collecting infrared cloud images of the front and back of the printed circuit board respectively; correcting each layer of infrared cloud image; merging the corrected front and back infrared cloud images into a double-channel infrared cloud image; inputting the reconstructed infrared cloud image into a double-channel multi-scale feature pyramid module to obtain component type and position information; using the component type and position information, and combining the thermal characteristics of the component, to further determine the state of the component, and to diagnose the fault of the double-layer printed circuit board under detection. The present application improves the double-layer printed circuit board component recognition capability using infrared cloud images; can solve the problem of mutual shielding of components on the printed circuit board, small geometric size and other defects that lead to undetectable problems; expands the range of printed circuit boards and the types of detectable faults, and improves the detection level of printed circuit boards.

Description

Technical Field

[0001] This invention relates to the field of printed circuit board testing technology, specifically to a method for identifying components on a two-layer printed circuit board using dual-channel infrared cloud images. Background Technology

[0002] Printed circuit boards (PCBs) are critical components in electronic products, enabling them to function properly. During the use of electronic products, PCBs can experience functional failures, primarily due to manufacturing defects and aging of components. Traditional invasive testing methods rely on technicians who measure electrical parameters such as voltage and current at key locations in the circuit, combining this information with circuit principles and experience to determine the location and type of PCB faults. This testing process requires a thorough understanding of the PCB schematic and design drawings, demanding significant prior knowledge and consuming considerable time and manpower. With increasing circuit integration, component sizes are shrinking, layouts are becoming denser, and the number of layers is constantly increasing. There is an urgent need for a non-invasive, more accurate, and faster method for fault detection in multilayer PCBs. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a method for identifying components on a double-layer printed circuit board using dual-channel infrared cloud images. The technical problem to be solved by this invention is achieved through the following technical solution:

[0004] A method for identifying components on a two-layer printed circuit board using dual-channel infrared cloud images, comprising the following steps:

[0005] Step 1: While keeping the printed circuit board in the same working state, collect infrared cloud images of arbitrary spatial coordinate angles on both sides of the printed circuit board.

[0006] Step 2: Correct the infrared cloud images of each layer. Use projection transformation to adjust two layers of infrared cloud images with different shapes, orientations and sizes acquired from the same printed circuit board to have the same shape, orientation and size.

[0007] Step 3: Combine the corrected front and back infrared cloud images into a single dual-channel infrared cloud image, and reconstruct the dual-channel infrared cloud image using equal-resolution raster interpolation. The resulting infrared cloud image serves as a sample for the subsequent feature extraction network.

[0008] Step 4: Input the reconstructed infrared cloud image into the dual-channel multi-scale feature pyramid module to obtain suggestion boxes and map the suggestion boxes onto the infrared feature map. Obtain the element type and location information through pre-class prediction and prediction box regression.

[0009] Step 5: Utilize the component type and location information, combined with the component's thermal characteristics, to determine the component's status and perform fault diagnosis on the tested double-layer printed circuit board.

[0010] The projection transformation in the second step includes the following steps:

[0011] Step 1: Let (u,v) be the imaging coordinates of the infrared thermal imager, and (x',y') be the coordinates after the center projection transformation. The corrected infrared cloud image size is H×W, satisfying the following mapping relationship:

[0012]

[0013] Step 2: Calculate the projection transformation of the coordinates using the following formula:

[0014]

[0015]

[0016] Where (x,y,z) are the transformed three-dimensional coordinates. For the affine transformation parameters, (a 31 a 32 ) represents the projection transformation parameters, (a 13 a 23 ) T Let a be the translation transformation parameter. 33 Let a be the scaling factor, and under normalization conditions, let a be the scaling factor. 33 =1.

[0017] The third step, performing equal-resolution raster interpolation reconstruction of the dual-channel infrared cloud image, includes the following steps:

[0018] Step 1: Calculate the scaling factors between the target image and the original image, including width, height, and orientation;

[0019] Step 2: Using the scaling factor, the virtual pixel positions in the original image are derived from the pixel positions in the target image.

[0020] Step 3: Find the four adjacent pixels in the width and height directions based on the virtual pixel position;

[0021] Step 4: Perform bilinear interpolation on the four pixels to obtain the pixel values ​​in the target image.

[0022] The component type and location information in step four are obtained through the following steps:

[0023] Step 1: Extract multi-scale infrared feature maps from the preprocessed infrared cloud images using the feature pyramid based on the Fast RCNN network;

[0024] Step 2: Obtain suggestion boxes using the regional suggestion network;

[0025] Step 3: Map the proposal box onto the infrared feature map using feature matching of interest to obtain the local common feature layer;

[0026] Step 4: Use the local common feature layer to perform category prediction and bounding box regression to obtain element type and location information.

[0027] The beneficial effects of this invention are: it improves the component identification capability of double-layer printed circuit boards using infrared cloud images; it can solve the problem of components on printed circuit boards being unable to be detected due to mutual obstruction and small geometric dimensions; it expands the scope of printed circuit boards and the types of detectable faults, and improves the detection level of printed circuit boards. Attached Figure Description

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] Figure 1 This is a schematic diagram illustrating the implementation steps of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be described more clearly and completely below with reference to the accompanying drawings in the embodiments. Of course, the described embodiments are only a part of the present invention and not all of it. Based on this embodiment, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of the present invention.

[0031] like Figure 1 As shown, a method for identifying components on a double-layer printed circuit board using dual-channel infrared cloud images includes the following steps:

[0032] Step 1: While keeping the printed circuit board in the same working state, collect infrared cloud images of arbitrary spatial coordinate angles on both sides of the printed circuit board.

[0033] The printed circuit board is placed stably on a horizontal surface and powered on for a period of time. At this time, the temperature of various parts of the printed circuit board reaches a steady state. The temperature distribution of the printed circuit board is collected by an infrared thermal imaging device to obtain the original infrared cloud image of the front.

[0034] The printed circuit board is inverted, and the temperature distribution of the printed circuit board is collected using an infrared thermal imaging device to obtain the original infrared cloud image of the back side.

[0035] Step 2: Correct the infrared cloud images of each layer. Use projection transformation to adjust two infrared cloud images of different shapes, orientations and sizes collected from the same printed circuit board to have the same shape, orientation and size. The imaging range of the infrared thermal imaging device not only covers the printed circuit board area, but also covers part of the area outside the printed circuit board. Before correcting the infrared cloud images of each layer, these areas need to be removed and only the printed circuit board area is retained.

[0036] The residual temperature of the environment around the printed circuit board is zero, but due to the heat generated by the components, there is a temperature rise in all parts of the printed circuit board, and the residual temperature is not zero. Therefore, by removing the areas with zero residual temperature in the infrared image, what remains is the printed circuit board area.

[0037] Two infrared cloud images with different shapes, orientations and sizes, acquired from the same printed circuit board, are adjusted to have the same shape, orientation and size to ensure that the two infrared cloud images can be correctly aligned and fused for analysis in the future. At the same time, the correction requires the use of projection transformation, but it is not necessary to perform affine transformation. The result after correction is not necessarily a rectangle.

[0038] Step 3: Combine the corrected front and back infrared cloud images into a single dual-channel infrared cloud image, and reconstruct the dual-channel infrared cloud image using equal-resolution raster interpolation. The resulting infrared cloud image serves as a sample for the subsequent feature extraction network.

[0039] Step 4: Input the reconstructed infrared cloud image into the dual-channel multi-scale feature pyramid module to obtain suggestion boxes and map the suggestion boxes onto the infrared feature map. Obtain the element type and location information through pre-class prediction and prediction box regression.

[0040] Step 5: Utilize the component type and location information, combined with the component's thermal characteristics, to determine the component's status and perform fault diagnosis on the tested double-layer printed circuit board.

[0041] The projection transformation in the second step includes the following steps:

[0042] Step 1: Let (u,v) be the imaging coordinates of the infrared thermal imager, and (x',y') be the coordinates after the center projection transformation. The corrected infrared cloud image size is H×W, satisfying the following mapping relationship:

[0043]

[0044] Step 2: Calculate the projection transformation of the coordinates using the following formula:

[0045]

[0046]

[0047] Where (x,y,z) are the transformed three-dimensional coordinates. For the affine transformation parameters, (a 31 a 32 ) represents the projection transformation parameters, (a 13 a 23 ) T Let a be the translation transformation parameter. 33 Let a be the scaling factor, and under normalization conditions, let a be the scaling factor. 33 =1.

[0048] The third step, performing equal-resolution raster interpolation reconstruction of the dual-channel infrared cloud image, includes the following steps:

[0049] Step 1: Calculate the scaling factors between the target image and the original image, including width, height, and orientation;

[0050] Step 2: Using the scaling factor, the virtual pixel positions in the original image are derived from the pixel positions in the target image.

[0051] Step 3: Find the four adjacent pixels in the width and height directions based on the virtual pixel position;

[0052] Step 4: Perform bilinear interpolation on the four pixels to obtain the pixel values ​​in the target image.

[0053] The component type and location information in step four are obtained through the following steps:

[0054] Step 1: Extract multi-scale infrared feature maps from the preprocessed infrared cloud map using the feature pyramid based on the Fast RCNN network. In practical applications, feature maps of three scales are extracted.

[0055] Step 2: Use the Region Proposal Network to obtain proposal boxes. Since the multi-scale feature pyramid enables the separate recognition of different scales, the number of prior boxes for each feature layer grid point in the Region Proposal Network is reduced from 9 to 3. Only 3 medium-sized prior boxes are retained, reducing the network complexity.

[0056] Step 3: Map the proposal boxes onto the infrared feature map using feature matching to obtain a local common feature layer; use RoIAlign instead of ROI pooling to improve the positional accuracy of the feature map mapping onto the infrared cloud map. RoIAlign cancels the quantization operation and uses bilinear interpolation to obtain the pixel values ​​with floating-point coordinates, eliminating the positional quantization error accumulated from the two operations of candidate box boundary quantization and average segmentation of the quantized region;

[0057] Step 4: Use the local common feature layer to perform category prediction and bounding box regression to obtain element type and location information.

[0058] The printed circuit board (PCB) inspection method provided by this invention relies on infrared cloud images as a single information source. It can automatically and accurately obtain the type and location information of all components from the infrared cloud images of a double-layer PCB, providing sufficient and reliable data for subsequent image segmentation, feature extraction, and fault diagnosis. Compared to existing automatic optical inspection methods, it eliminates the need for optical image acquisition, overcoming the drawback of optical image quality depending on imaging conditions. Simultaneously, it enables the detection of optically invisible defects, expanding the scope of inspection objects and the types of detectable faults, thus improving the overall inspection level of PCBs.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely prisms of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying components on a double-layer printed circuit board using dual-channel infrared cloud images, characterized in that: The method includes the following steps: Step 1: While keeping the printed circuit board in the same working state, collect infrared cloud images of arbitrary spatial coordinate angles on both sides of the printed circuit board. Step 2: Correct the infrared cloud images of each layer. Use projection transformation to adjust two layers of infrared cloud images with different shapes, orientations and sizes acquired from the same printed circuit board to have the same shape, orientation and size. Step 3: Combine the corrected front and back infrared cloud images into a single dual-channel infrared cloud image, and reconstruct the dual-channel infrared cloud image using equal-resolution raster interpolation. The resulting infrared cloud image serves as a sample for the subsequent feature extraction network. Step 4: Input the reconstructed infrared cloud image into the dual-channel multi-scale feature pyramid module to obtain suggestion boxes and map the suggestion boxes onto the infrared feature map. Obtain the element type and location information through pre-class prediction and prediction box regression. Step 5: Utilize the component type and location information, combined with the component's thermal characteristics, to determine the component's condition and perform fault diagnosis on the tested double-layer printed circuit board; The component type and location information in step four are obtained through the following steps: Step 1: Extract multi-scale infrared feature maps from the preprocessed infrared cloud images using the feature pyramid based on the Fast RCNN network; Step 2: Obtain suggestion boxes using the regional suggestion network; Step 3: Map the proposal box onto the infrared feature map using feature matching of interest to obtain the local common feature layer; Step 4: Use the local common feature layer to perform category prediction and bounding box regression to obtain element type and location information.

2. The method for identifying components on a double-layer printed circuit board using dual-channel infrared cloud images according to claim 1, characterized in that: The projection transformation in the second step includes the following steps: Step 1: Set The imaging coordinates of the infrared thermal imager, The coordinates after the center projection transformation are given, and the corrected infrared cloud image size is [value missing]. The following mapping relationship is satisfied: ; Step 2: Calculate the projection transformation of the coordinates using the following formula: ; in, The transformed three-dimensional coordinates For affine transformation parameters, These are the projection transformation parameters. These are the parameters for the translation transformation. Let be the scaling factor, and under normalization conditions, take . .

3. The method for identifying components on a double-layer printed circuit board using dual-channel infrared cloud images according to claim 1, characterized in that: In the third step, the dual-channel infrared cloud image is reconstructed using equal-resolution raster interpolation. Includes the following steps: Step 1: Calculate the scaling factors between the target image and the original image, including width, height, and orientation; Step 2: Using the scaling factor, the virtual pixel positions in the original image are derived from the pixel positions in the target image. Step 3: Find the four adjacent pixels in the width and height directions based on the virtual pixel position; Step 4: Perform bilinear interpolation on the four pixels to obtain the pixel values ​​in the target image.

Images