A method and device for reading an instrument of a pointer pressure gauge

Abstract

The present invention discloses an instrument reading method and device for a pointer pressure gauge. The method comprises the steps of: obtaining a first pressure gauge image of a pointer pressure gauge taken by an inspection robot, segmenting the first pressure gauge image to obtain a second pressure gauge image containing only the pressure gauge; correcting the second pressure gauge image to a third pressure gauge image at a preset angle by perspective transformation, wherein the preset angle is the angle at which the camera is facing the dial; inputting the third pressure gauge image into a pressure gauge classification model based on a deep learning neural network to obtain the reading of the pointer pressure gauge. The technical solution of the present invention improves the accuracy of the inspection robot in reading the pointer pressure gauge.

Description

Technical Field

[0001] The invention relates to the technical field of pointer type pressure gauge reading, and in particular to an instrument reading method and device applied to a pointer type pressure gauge. Background Art

[0002] The present invention is applicable to the field of patrol robots. There are numerous pointer pressure gauges distributed in every corner of the factory. These pressure gauges are used to record various indicators. The readings displayed by the pressure gauges are crucial to the safety and normal operation of the factory. Pressure gauges will be installed in every corner of the factory. Manually counting the pressure gauges will bring certain labor costs to the factory. In addition, some corners are inconvenient for people to move around. Therefore, it is very necessary to use patrol robots to realize the automatic reading of pressure gauges. There are many pointer pressure gauges distributed in various factories. The pointer pressure gauges are used to indicate whether the factory is running normally or not. The reading of the pressure gauge is crucial. At present, the most reading methods used in factories are manual readings, which rely on patrol personnel to read and record the pressure gauges. In view of this demand for pressure gauge readings in large factories, patrol robots can just solve this problem. The patrol robot can walk freely in the factory, use a camera to detect the position of the pressure gauge, and can read the pointer pressure gauge, complete the record, and transmit the reading to the background.

[0003] At present, the reading of pointer pressure gauges is mainly based on traditional machine learning methods, which mainly include the following steps:

[0004] 1. Dial extraction: Preprocess the image containing the instrument panel, crop the dial, and remove the background, mainly using mean filtering, grayscale conversion, and probabilistic Hough circle detection.

[0005] 2. Scale line extraction: Through contour search, all black areas (scale lines, pointers, interference points) can be found, and then the scale line contour is fitted with a straight line to find the center of the dial.

[0006] 3. Pointer outline extraction: After removing the scale lines and noise points in the original image, the remaining outline only contains the pointer and the disk.

[0007] 4. Calculate the reading based on the reading between the pointer and the zero scale.

[0008] The reading method of the prior art has many steps, and there will be many interference points when machine learning performs scale line detection and pointer detection. When the light, shooting angle and the zero scale of the pressure gauge are not in a fixed position, the reading of the prior art will have a large deviation. Summary of the invention

[0009] The present invention provides an instrument reading method and device applied to a pointer type pressure gauge, which improves the accuracy of the inspection robot in reading the pointer type pressure gauge.

[0010] An embodiment of the present invention provides an instrument reading method applied to a pointer pressure gauge, comprising the following steps:

[0011] Acquire a first pressure gauge image of a pointer-type pressure gauge photographed by the inspection robot, and obtain a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image;

[0012] Correcting the second pressure gauge image into a third pressure gauge image at a preset angle by perspective transformation, where the preset angle is the angle at which the camera directly faces the dial;

[0013] The third pressure gauge image is input into a pressure gauge classification model based on a deep learning neural network to obtain a reading of the pointer pressure gauge.

[0014] Further, according to the following steps, a training set of the pressure gauge classification model is obtained:

[0015] Acquire a fourth pressure gauge image of the pointer-type pressure gauge photographed by the inspection robot, and obtain a fifth pressure gauge image containing only the pressure gauge after segmenting the fourth pressure gauge image;

[0016] Correcting the fifth pressure gauge image into a sixth pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial;

[0017] Separating the dial and the pointer of the sixth pressure gauge image to obtain a pointer image containing only the pointer and a dial image containing only the dial;

[0018] A seventh pressure gauge image with multiple angle combinations is generated according to the pointer image and the dial image, and the seventh pressure gauge image is used as a training set for the pressure gauge classification model; the multiple angle combinations refer to image combinations of multiple pointer angles and multiple dial angles.

[0019] Furthermore, the multiple pointer angles include 360 ​​types, the multiple dial angles include 360 ​​types, the pointer angle refers to the angle between the pointer in the image and the vertical direction, and the dial angle refers to the angle between the zero scale line of the dial in the image and the vertical direction.

[0020] Furthermore, the second pressure gauge image is corrected into a third pressure gauge image at a preset angle by perspective transformation, specifically:

[0021] Fit the elliptical boundary of the pressure gauge in the third pressure gauge image, obtain the four vertices corresponding to the major axis and the minor axis of the ellipse, use OpenCV to perspective transform the four vertices of the ellipse into four vertices of a circle, and obtain the third pressure gauge image corrected to a preset angle.

[0022] Furthermore, the process of obtaining the prediction sequence is as follows:

[0023] The sequences of the collected text files are annotated using an extended label set to obtain a first sequence set, and the sequences in the first sequence set that can be converted into real sequences through a mapping function are determined as predicted sequences.

[0024] Furthermore, when generating the reply text, the natural language generation module includes the following steps:

[0025] Determine the information that needs to be replied according to the received semantic analysis result, and determine a reasonable text order according to the information that needs to be replied;

[0026] Determining text information presented in a single sentence according to the text sequence, and selecting a corresponding plurality of words and phrases according to the text information of the single sentence;

[0027] Identify the field to which the information to be replied belongs, and select words and phrases corresponding to the field from the plurality of words and phrases according to the field;

[0028] The selected words and phrases of the corresponding field are combined into sentences with correct format.

[0029] Furthermore, the training process of the pressure gauge classification model includes the following steps:

[0030] Inputting the seventh pressure gauge image into the pressure gauge classification model, extracting image features through a 3×3 convolution layer to generate a first feature map;

[0031] Performing image feature extraction on the first feature map through the MBconv module to obtain a fourth feature map;

[0032] Performing a dimensionality reduction operation on the fourth feature map through a 1×1 convolution layer, and then inputting the result to a pooling layer and a fully connected layer to obtain a classification result of the seventh pressure gauge image;

[0033] After calculating the loss according to the classification result and the original label of the seventh pressure gauge image, back propagation is performed to update and iterate the parameters of the pressure gauge classification model to obtain the converged pressure gauge classification model.

[0034] Further, the first feature map is subjected to image feature extraction by the MBconv module to obtain a fourth feature map, specifically:

[0035] The first feature map is dimensionally increased through a 1×1 convolution layer, each channel of the dimensionally increased first feature map is convolved through a Depthwise Conv convolution, and then the output of each channel is concatenated to extract image features to obtain a second feature map;

[0036] After the extracted second feature map is input into the SE module, the second feature map is input into a 1×1 convolutional layer for dimensionality reduction, and then outputs a third feature map through a Dropout layer;

[0037] After the third feature map is fused with the second feature map of the SE module, a fourth feature map is output.

[0038] Another embodiment of the present invention provides an instrument reading device applied to a pointer-type pressure gauge, including a pressure gauge image acquisition module, a pressure gauge image correction module and a pressure gauge reading module.

[0039] The pressure gauge image acquisition module is used to acquire a first pressure gauge image of a pointer-type pressure gauge photographed by the inspection robot, and obtain a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image;

[0040] The pressure gauge image correction module is used to correct the second pressure gauge image into a third pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial;

[0041] The pressure gauge reading module is used to input the third pressure gauge image into a pressure gauge classification model based on a deep learning neural network to obtain the reading of the pointer pressure gauge.

[0042] The embodiments of the present invention have the following beneficial effects:

[0043] The present invention provides an instrument reading method and device for a pointer pressure gauge. The present invention obtains a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image taken by the inspection robot, corrects the second pressure gauge image to a third pressure gauge image at a preset angle through perspective transformation, and the preset angle is the angle at which the camera directly faces the dial. The third pressure gauge image is input into a pressure gauge classification model based on a deep learning neural network, and an accurate reading of the pointer pressure gauge can be obtained. The present invention performs segmentation and correction processing before inputting the pressure gauge image into the model, so that the image input into the model is more accurate and has a higher degree of recognition, thereby improving the accuracy of the model reading the pressure gauge. At the same time, when training the pressure gauge classification model based on a deep learning neural network of the present invention, a seventh pressure gauge image with a combination of multiple angles is generated according to the pointer image and the dial image, and the seventh pressure gauge image is used as the training set of the pressure gauge classification model, thereby making the training accuracy of the model higher, and further improving the accuracy of reading the pointer pressure gauge. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of an instrument reading method applied to a pointer pressure gauge provided by an embodiment of the present invention;

[0045] Figure 2 It is a structural schematic diagram of an instrument reading device applied to a pointer-type pressure gauge provided by one embodiment of the present invention;

[0046] Figure 3 A pointer image including only a pointer in an instrument reading method for a pointer pressure gauge provided by an embodiment of the present invention;

[0047] Figure 4 A dial image including only a dial in a method for reading an instrument applied to a pointer pressure gauge provided by an embodiment of the present invention;

[0048] Figure 5 is a seventh pressure gauge image of an instrument reading method applied to a pointer-type pressure gauge provided by an embodiment of the present invention;

[0049] Figure 6 It is a structural schematic diagram of an MBconv module of an instrument reading method applied to a pointer pressure gauge provided by an embodiment of the present invention;

[0050] Figure 7 It is a structural schematic diagram of an SE module of an instrument reading method applied to a pointer pressure gauge provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, an instrument reading method applied to a pointer pressure gauge provided by an embodiment of the present invention comprises the following steps:

[0053] A first pressure gauge image of a pointer-type pressure gauge photographed by an inspection robot is acquired, and a second pressure gauge image containing only the pressure gauge is obtained after segmentation processing of the first pressure gauge image; preferably, the first pressure gauge image is input into a segmentation network model to obtain a second pressure gauge image containing only the pressure gauge.

[0054] The second pressure gauge image is corrected into a third pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial.

[0055] The third pressure gauge image is input into a pressure gauge classification model based on a deep learning neural network to obtain a reading of the pointer pressure gauge.

[0056] As one embodiment, the second pressure gauge image is corrected into a third pressure gauge image at a preset angle by perspective transformation, specifically:

[0057] Fit the elliptical boundary of the pressure gauge in the third pressure gauge image, obtain the four vertices corresponding to the major axis and the minor axis of the ellipse, use OpenCV to perspective transform the four vertices of the ellipse into four vertices of a circle, and obtain the third pressure gauge image corrected to a preset angle.

[0058] As one embodiment, a training set of the pressure gauge classification model is obtained according to the following steps:

[0059] Acquire a fourth pressure gauge image of the pointer-type pressure gauge photographed by the inspection robot, and obtain a fifth pressure gauge image containing only the pressure gauge after segmenting the fourth pressure gauge image;

[0060] Correcting the fifth pressure gauge image to a sixth pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial; that is, adjusting the fifth pressure gauge image to an image obtained when the camera is flush with the pressure gauge and facing the pressure gauge;

[0061] like Figure 3 and Figure 4As shown, the dial and the pointer of the sixth pressure gauge image are separated to obtain a pointer image containing only the pointer and a dial image containing only the dial;

[0062] A seventh pressure gauge image with multiple angle combinations is generated according to the pointer image and the dial image, such as Figure 5 As shown, the seventh pressure gauge image is used as the training set of the pressure gauge classification model; the multiple angle combinations refer to image combinations of multiple pointer angles and multiple dial angles. Specifically, the pointer image and the dial image are converted into the seventh pressure gauge image of multiple angle combinations through an algorithm.

[0063] The multiple pointer angles include 360 ​​types, and the multiple dial angles include 360 ​​types. The pointer angle refers to the angle between the pointer in the image and the vertical direction, and the dial angle refers to the angle between the zero scale line of the dial in the image and the vertical direction. Specifically, according to the angle range of 0° to 359° (the angle value is an integer), the pointer angle and the dial angle are divided into 360 types. The seventh pressure gauge image is divided into 360 types according to the angle between the pointer in the image and the vertical direction.

[0064] As one embodiment, the training process of the pressure gauge classification model includes the following steps:

[0065] Inputting the seventh pressure gauge image into the pressure gauge classification model, extracting image features through a 3×3 convolution layer to generate a first feature map;

[0066] Performing image feature extraction on the first feature map through the MBconv module to obtain a fourth feature map;

[0067] Performing a dimensionality reduction operation on the fourth feature map through a 1×1 convolution layer, and then inputting the result to a pooling layer and a fully connected layer to obtain a classification result of the seventh pressure gauge image;

[0068] After calculating the loss according to the classification result and the original label of the seventh pressure gauge image, back propagation is performed to update and iterate the parameters of the pressure gauge classification model to obtain the converged pressure gauge classification model.

[0069] As one example, Figure 6As shown in the figure, the MBconv module includes a 1×1 convolution layer (the convolution layer plays a role of dimensionality increase, including Batch Normalization and Swish activation function), a k×k DepthwiseConv convolution (including Batch Normalization and Swish activation function), an SE module (i.e., Squeeze-and-Excitation module), a 1×1 convolution layer (the convolution layer plays a role of dimensionality reduction, including BatchNormalization) and a Dropout layer.

[0070] The MBconv module is used to extract image features from the first feature map to obtain a fourth feature map, specifically:

[0071] The first feature map is dimensionally increased through a 1×1 convolution layer, each channel of the dimensionally increased first feature map is convolved through a Depthwise Conv convolution, and then the output of each channel is concatenated to extract image features to obtain a second feature map;

[0072] After the extracted second feature map is input into the SE module, the second feature map is input into the 1×1 convolution layer for dimensionality reduction, and then the third feature map is output through the Dropout layer; Figure 7 As shown, the SE (Squeeze-and-Excitation) module includes an average pooling layer AvgPooling and two fully connected layers FC1 and FC2.

[0073] After the third feature map is fused with the second feature map of the SE module, a fourth feature map is output.

[0074] The present invention obtains a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image taken by the inspection robot, corrects the second pressure gauge image to a third pressure gauge image of a preset angle through perspective transformation, and the preset angle is the angle at which the camera directly faces the dial. The third pressure gauge image is input into a pressure gauge classification model based on a deep learning neural network, and an accurate reading of the pointer pressure gauge can be obtained. The present invention performs segmentation and correction processing before inputting the pressure gauge image into the model, so that the image input into the model is more accurate and has a higher degree of recognition, thereby improving the accuracy of the model reading the pressure gauge. At the same time, when training the pressure gauge classification model based on a deep learning neural network of the present invention, a seventh pressure gauge image of a combination of multiple angles is generated according to the pointer image and the dial image, and the seventh pressure gauge image is used as the training set of the pressure gauge classification model, thereby making the training accuracy of the model higher, and further improving the accuracy of reading the pointer pressure gauge.

[0075] like Figure 2 As shown, another embodiment of the present invention provides an instrument reading device for a pointer-type pressure gauge, including a pressure gauge image acquisition module, a pressure gauge image correction module and a pressure gauge reading module.

[0076] The pressure gauge image acquisition module is used to acquire a first pressure gauge image of a pointer-type pressure gauge photographed by the inspection robot, and obtain a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image;

[0077] The pressure gauge image correction module is used to correct the second pressure gauge image into a third pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial;

[0078] The pressure gauge reading module is used to input the third pressure gauge image into a pressure gauge classification model based on a deep learning neural network to obtain the reading of the pointer pressure gauge.

[0079] For the convenience and brevity of description, the instrument reading device applied to a pointer pressure gauge of the device item embodiment of the present invention includes all the implementation methods of the above-mentioned instrument reading method embodiment applied to a pointer pressure gauge, which will not be repeated here.

[0080] It should be noted that the device embodiments described above are only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement it without paying creative labor. The memory can be used to store the computer program and / or module, and the processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as a hard disk, an internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0081] Those of ordinary skill in the art can understand and implement it without creative work. The above is a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention. Those of ordinary skill in the art can understand that all or part of the processes in the above embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above embodiments. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

Claims

1. An instrument reading method applied to a pointer pressure gauge, characterized in that: The following steps are involved: Acquire a first pressure gauge image of a pointer-type pressure gauge photographed by the inspection robot, and obtain a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image; Correcting the second pressure gauge image into a third pressure gauge image at a preset angle by perspective transformation, where the preset angle is the angle at which the camera directly faces the dial; Inputting the third pressure gauge image into a pressure gauge classification model based on a deep learning neural network to obtain a reading of the pointer pressure gauge; According to the following steps, a training set of the pressure gauge classification model is obtained: Acquire a fourth pressure gauge image of the pointer-type pressure gauge photographed by the inspection robot, and obtain a fifth pressure gauge image containing only the pressure gauge after segmenting the fourth pressure gauge image; Correcting the fifth pressure gauge image into a sixth pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial; Separating the dial and the pointer of the sixth pressure gauge image to obtain a pointer image containing only the pointer and a dial image containing only the dial; generating a seventh pressure gauge image of multiple angle combinations according to the pointer image and the dial image, and using the seventh pressure gauge image as a training set for the pressure gauge classification model; the multiple angle combinations refer to image combinations of multiple pointer angles and multiple dial angles; The multiple pointer angles include 360 ​​kinds, and the multiple dial angles include 360 ​​kinds. The pointer angle refers to the angle between the pointer in the image and the vertical direction, and the dial angle refers to the angle between the zero scale line of the dial in the image and the vertical direction; The training process of the pressure gauge classification model includes the following steps: Inputting the seventh pressure gauge image into the pressure gauge classification model, extracting image features through a 3×3 convolution layer to generate a first feature map; Performing image feature extraction on the first feature map through the MBconv module to obtain a fourth feature map; Performing a dimensionality reduction operation on the fourth feature map through a 1×1 convolution layer, and then inputting the result to a pooling layer and a fully connected layer to obtain a classification result of the seventh pressure gauge image; After calculating the loss according to the classification result and the original label of the seventh pressure gauge image, back propagation is performed to update and iterate the parameters of the pressure gauge classification model to obtain the converged pressure gauge classification model; the image feature extraction of the first feature map is performed through the MBconv module to obtain the fourth feature map, which is specifically: The first feature map is dimensionally increased through a 1×1 convolution layer, each channel of the dimensionally increased first feature map is convolved through a Depthwise Conv convolution, and then the output of each channel is concatenated to extract image features to obtain a second feature map; After the extracted second feature map is input into the SE module, the second feature map is input into a 1×1 convolutional layer for dimensionality reduction, and then outputs a third feature map through a Dropout layer; After the third feature map is fused with the second feature map of the SE module, a fourth feature map is output.

2. The instrument reading method for a pointer pressure gauge according to claim 1, characterized in that: The second pressure gauge image is corrected into a third pressure gauge image at a preset angle by perspective transformation, specifically: Fit the elliptical boundary of the pressure gauge in the third pressure gauge image, obtain the four vertices corresponding to the major axis and the minor axis of the ellipse, use OpenCV to perspective transform the four vertices of the ellipse into four vertices of a circle, and obtain the third pressure gauge image corrected to a preset angle.

3. An instrument reading device for a pointer pressure gauge, characterized in that: It includes a pressure gauge image acquisition module, a pressure gauge image correction module and a pressure gauge reading module; The pressure gauge image acquisition module is used to acquire a first pressure gauge image of a pointer-type pressure gauge photographed by the inspection robot, and obtain a second pressure gauge image containing only the pressure gauge after segmenting the first pressure gauge image; The pressure gauge image correction module is used to correct the second pressure gauge image into a third pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial; The pressure gauge reading module is used to input the third pressure gauge image into a pressure gauge classification model based on a deep learning neural network to obtain the reading of the pointer pressure gauge; According to the following steps, a training set of the pressure gauge classification model is obtained: Acquire a fourth pressure gauge image of the pointer-type pressure gauge photographed by the inspection robot, and obtain a fifth pressure gauge image containing only the pressure gauge after segmenting the fourth pressure gauge image; Correcting the fifth pressure gauge image into a sixth pressure gauge image at a preset angle through perspective transformation, where the preset angle is the angle at which the camera is facing the dial; Separating the dial and the pointer of the sixth pressure gauge image to obtain a pointer image containing only the pointer and a dial image containing only the dial; generating a seventh pressure gauge image of multiple angle combinations according to the pointer image and the dial image, and using the seventh pressure gauge image as a training set for the pressure gauge classification model; the multiple angle combinations refer to image combinations of multiple pointer angles and multiple dial angles; The multiple pointer angles include 360 ​​kinds, and the multiple dial angles include 360 ​​kinds. The pointer angle refers to the angle between the pointer in the image and the vertical direction, and the dial angle refers to the angle between the zero scale line of the dial in the image and the vertical direction; The training process of the pressure gauge classification model includes the following steps: Inputting the seventh pressure gauge image into the pressure gauge classification model, extracting image features through a 3×3 convolution layer to generate a first feature map; Performing image feature extraction on the first feature map through the MBconv module to obtain a fourth feature map; Performing a dimensionality reduction operation on the fourth feature map through a 1×1 convolution layer, and then inputting the result to a pooling layer and a fully connected layer to obtain a classification result of the seventh pressure gauge image; After calculating the loss according to the classification result and the original label of the seventh pressure gauge image, back propagation is performed to update and iterate the parameters of the pressure gauge classification model to obtain the converged pressure gauge classification model; The MBconv module is used to extract image features from the first feature map to obtain a fourth feature map, specifically: The first feature map is dimensionally increased through a 1×1 convolution layer, each channel of the dimensionally increased first feature map is convolved through a Depthwise Conv convolution, and then the output of each channel is concatenated to extract image features to obtain a second feature map; After the extracted second feature map is input into the SE module, the second feature map is input into a 1×1 convolutional layer for dimensionality reduction, and then outputs a third feature map through a Dropout layer; After the third feature map is fused with the second feature map of the SE module, a fourth feature map is output.

Images