Optical phased array main lobe centroid recognition method and system based on CBAM-CNN

Abstract

The application discloses a kind of optical phased array main lobe centroid identification method and system based on CBAM-CNN, its identification method includes the following steps: S1 adjusts optical antenna unit, S2 adjusts image acquisition unit hardware exposure parameter, S3 converts gray value image into binary image, S4 binary image is input based on CBAM module convolutional neural network, S5 method judges the centroid position of main lobe area light spot, outputs main lobe area centroid coordinate, S6 judges whether optical phased array far-field light spot exceeds image acquisition unit hardware collection field of view.The convolutional neural network of the application combines channel attention mechanism and spatial attention mechanism, accurately image area identification is carried out to optical phased array main lobe area, optical phased array side lobe area and stray light area, multi-spot identification problem is converted into single-spot identification problem in optical phased array main lobe area, and optical phased array main lobe centroid can be accurately identified in the case of multiple side lobe interference.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of space laser communication, and particularly relates to a main lobe centroid identification method and system for an optical phased array based on CBAM-CNN. BACKGROUND

[0002] The optical phased array technology has the advantages of fast response speed, high pointing accuracy, light weight, and small size, and has great development potential in the fields of free space laser communication, solid state laser radar, and long distance three-dimensional imaging.

[0003] The existing spot centroid identification method mainly relies on image gray value threshold related algorithm to locate the spot centroid position. For example, the invention patent with the publication number CN111462225 A discloses a centroid identification and positioning method for an infrared spot image, which comprises: taking an image of an infrared spot centroid region to be identified and positioned as a region of interest image; performing deep conversion and gray linear stretching on the region of interest image to generate a preprocessed image; performing adaptive threshold calculation on the preprocessed image to obtain a binary threshold value of the preprocessed image; binarizing the preprocessed image, and finding out the object contour in the region according to the contour tracking algorithm, and selecting the maximum contour point set as the spot contour; and positioning the infrared spot centroid according to the spot contour. The invention patent with the publication number CN100371676C discloses a fast and high-precision spot image centroid positioning method and device, which comprises: performing Gaussian convolution operation on the pixel gray value, and judging whether the pixel gray value after Gaussian convolution operation is greater than a preset threshold, if yes, marking the current pixel, identifying the belonging spot, and calculating the product of the current pixel gray value and coordinate value and the accumulated value of all pixel gray values and coordinate values of the same spot that have been processed, the accumulated value of the gray value of the current pixel and all pixel gray values of the same spot that have been processed, and saving the obtained accumulated value, otherwise, marking the current pixel as a background pixel and processing; after processing the entire output image, calculating the quotient of the accumulated value of the product of the gray value and the coordinate value and the accumulated value of the gray value, and outputting the calculation result as the spot image centroid coordinate value. The invention also discloses a centroid positioning device. The invention can improve the data processing speed and noise immunity in spot image centroid positioning, and can process multiple spot images.

[0004] However, the spot formed by the optical phased array exit beam in the far field includes a main lobe and a side lobe. The existence of the optical phased array side lobe and stray light in the optical system will interfere with the centroid identification of the far field spot, resulting in that the existing image gray value threshold related positioning algorithm cannot accurately identify the optical phased array far field spot. Therefore, a method capable of accurately identifying the main lobe centroid of the optical phased array in the presence of multiple side lobes and stray light interference has become a problem to be solved in the engineering application of the optical phased array technology. SUMMARY

[0005] The application aims to provide a CBAM-CNN-based optical phased array main lobe centroid identification method and system.

[0006] To achieve the above-mentioned purposes, the technical scheme adopted by the application is as follows:

[0007] A CBAM-CNN-based optical phased array main lobe centroid identification method, comprising the following steps:

[0008] S1, adjusting the optical antenna unit so that the optical phased array far-field spot enters the field of view range of the image acquisition unit, and adjusting the optical antenna unit so that the main lobe of the optical phased array far-field spot appears at the center of the camera field of view range;

[0009] S2, adjusting the hardware exposure parameters of the image acquisition unit so that as many spots as possible are displayed in its field of view without excessive exposure;

[0010] S3, the image processing unit reads the current gray value image of the image acquisition unit, and converts the collected gray value image into a binary image by setting a gray value threshold;

[0011] S4, the image processing unit inputs the obtained binary image into a CBAM module convolutional neural network, and the CBAM module convolutional neural network outputs the main lobe region of the optical phased array, the sidelobe region of the optical phased array, and the stray light region;

[0012] S5, the centroid identification unit determines the centroid position of the main lobe region spot by using the centroid method, and outputs the centroid coordinates of the main lobe region of the optical phased array far-field spot;

[0013] S6, the negative feedback adjustment unit determines whether the optical phased array far-field spot exceeds the hardware acquisition field of view of the image acquisition unit, if it exceeds the field of view, it jumps to step S1, if it does not exceed the field of view, it jumps to step S3.

[0014] As a further optimization of the above method,

[0015] Preferably, in step S4, the identification of the main lobe region and the sidelobe region by the CBAM module convolutional neural network can be realized by the following steps:

[0016] S41, obtaining an optical phased array far-field spot binary image dataset for training a convolutional neural network;

[0017] S42, establishing a convolutional neural network based on a CBAM module, training the neural network using the optical phased array far-field spot binary image dataset to obtain a trained convolutional neural network based on the CBAM module;

[0018] S43, inputting an optical phased array far-field spot binary image to be identified into the trained convolutional neural network based on the CBAM module.

[0019] Further, the CBAM-based convolutional neural network in step S42 is a multi-layer convolutional neural network, the first, third and fifth layers of the multi-layer convolutional neural network are convolutional layers, the second, fourth and sixth layers of the multi-layer convolutional neural network are CBAM layers, the seventh layer of the multi-layer convolutional neural network is a fully connected layer, and the eighth layer of the multi-layer convolutional neural network is a Softmax layer.

[0020] The CBAM layer is composed of a channel attention module and a spatial attention module.

[0021] The corresponding process description formula of the channel attention module is:

[0022]

[0023] In the formula, is the channel attention module weight coefficient of the intermediate feature F, is a Sigmoid function, MLP is a multi-layer neural network, AvgPool is an average pooling, and MaxPool is a maximum pooling, and is the weight of the multi-layer neural network MLP, is the average pooling value of the intermediate feature F in the channel attention module, is the maximum pooling value of the intermediate feature F in the channel attention module.

[0024] The corresponding process description formula of the spatial attention module is:

[0025]

[0026] In the formula, is the spatial attention module weight coefficient of the intermediate feature F, is a convolution kernel with a size of 5x5, is the average pooling value of the intermediate feature F in the spatial attention module, is the maximum pooling value of the intermediate feature F in the spatial attention module.

[0027] The CBAM layer, the corresponding process description formula is:

[0028]

[0029] In the formula, is the intermediate quantity of the intermediate feature F through the channel attention module, is the output quantity of the intermediate feature F through the channel attention module and the spatial attention module.

[0030] An optical phased array main lobe centroid recognition system based on a CBAM module convolutional neural network, comprising an optical antenna unit, an image acquisition unit, an image processing unit, a centroid recognition unit, and a negative feedback adjustment unit.

[0031] The optical antenna unit is used to receive a target light beam and transmit the light beam to the image acquisition unit. The relative position of the target light beam irradiated on the image acquisition unit can be changed by adjusting the optical antenna unit.

[0032] The image acquisition unit is used to acquire a light spot image generated by the target light beam irradiated thereon. The light spot brightness information of the acquired image can be changed by changing the hardware exposure time and other parameters of the image acquisition unit.

[0033] The image processing unit is used to convert the acquired light spot gray value image into a binary image, and realize image region recognition of the main lobe region of the optical phased array, the sidelobe region of the optical phased array, and the stray light region through a convolutional neural network based on a CBAM module.

[0034] The centroid recognition unit uses the centroid method to determine the centroid position of the light spot in the main lobe region of the optical phased array, and outputs the centroid coordinates of the main lobe region of the optical phased array far-field light spot.

[0035] The negative feedback adjustment unit is used to compare the deviation of the centroid coordinates of the main lobe region of the optical phased array far-field light spot and the field center coordinates of the image acquisition unit, calculate the related control parameters of the optical antenna unit, and control the optical antenna unit to move the light spot centroid to the field center of the image acquisition unit.

[0036] Preferably, the image processing unit adopts a supervised machine learning method in the region recognition based on the CBAM module convolutional neural network target detection method, and the training method is as follows:

[0037] (1) Establish a data set

[0038] Collecting a binary image containing an optical phased array main lobe, an optical phased array side lobe and a stray light spot and establishing a data set;

[0039] (2) Image annotation

[0040] The data set is manually annotated to mark the optical phased array main lobe region, the optical phased array side lobe region, the stray light region, and the data set of the light spot type and the light spot position is used for subsequent model training and evaluation;

[0041] (3) Model training

[0042] The CBAM module-based convolutional neural network is a multi-layer convolutional neural network, the first, third and fifth layers in the multi-layer convolutional neural network are convolutional layers, the second, fourth and sixth layers in the multi-layer convolutional neural network are CBAM layers, the seventh layer in the multi-layer convolutional neural network is a fully connected layer, and the eighth layer in the multi-layer convolutional neural network is a Softmax layer.

[0043] The CBAM module enhances the extraction ability of the convolutional neural network for different channel features and different position features by introducing a channel attention module and a spatial attention module.

[0044] (4) Result evaluation

[0045] The mean average precision mAP is used to evaluate the accuracy of the model prediction result, and the calculation formula is as follows:

[0046]

[0047] Where C represents the total number of categories, The AP value of the i-th target.

[0048] Preferably, steps S1 to S6 are executed in a loop, and the optical phased array far-field light spot main lobe region centroid coordinates are output once for each execution round, and the loop period ends when the number of cycles reaches a predetermined value.

[0049] Compared with the prior art, the present application has the following advantages:

[0050] (1) The convolutional neural network of the present application combines channel attention mechanism and spatial attention mechanism, accurately identifies the image region for the optical phased array main lobe region, the optical phased array side lobe region and the stray light region, converts the multi-spot identification problem into a single-spot identification problem in the optical phased array main lobe region, and can accurately identify the optical phased array main lobe centroid in the presence of multiple side lobe interference, eliminate multiple side lobe interference, and only calculate the centroid coordinates of the main lobe spot, solving the problem of accurately identifying the optical phased array main lobe centroid in the presence of multiple side lobe interference.

[0051] (2) By identifying the region of stray light, the present invention can reduce the interference of stray light on the identification of the centroid of the spot, thereby further improving the accuracy of the identification of the centroid of the spot.

[0052] (3) The present invention introduces a negative feedback adjustment unit, which controls the optical antenna unit to move the centroid of the spot to the center of the field of view of the image acquisition unit by comparing the deviation between the centroid coordinates of the main lobe region of the optical phased array far field spot and the center coordinates of the field of view of the image acquisition unit; so that the main lobe of the spot is always near the center of the field of view, ensuring the accuracy of the centroid recognition of the optical phased array spot.

[0053] (4) This invention addresses the problem of identifying the centroid of multiple light spots formed by the linearly symmetrical arrangement of the main lobe and multiple side lobes of the beam emitted from the optical phased array in the far field. This solution can be used for identifying the centroid of multiple light spots that satisfy the requirement that the side lobes are symmetrically distributed relative to the center of the main lobe. It is independent of the specific technical system adopted by the optical phased array and can be applied to optical phased array devices under various technical systems such as optical waveguide optical phased array and liquid crystal optical phased array. It has a wide range of applications. Attached Figure Description

[0054] Figure 1 This is a block diagram of a main lobe centroid identification system for an optical phased array based on a CBAM module convolutional neural network according to the present invention.

[0055] Figure 2 This is a schematic diagram of image partitioning in the field of view of the image acquisition unit of the present invention;

[0056] Figure 3 This is a schematic diagram of the channel attention module structure of the present invention;

[0057] Figure 4 This is a schematic diagram of the spatial attention module structure of the present invention;

[0058] Figure 5 This is a schematic diagram of the CBAM module structure of the present invention;

[0059] Figure 6 This is a flowchart illustrating an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives and advantages of the present invention clearer, the present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0061] like Figure 1 As shown, this invention discloses a main lobe centroid recognition system based on CBAM-CNN for optical phased arrays, including an optical antenna unit, an image acquisition unit, an image processing unit, a centroid recognition unit, and a negative feedback adjustment unit.

[0062] The optical antenna unit is responsible for receiving the target light beam and transmitting the light beam onto the image acquisition unit. The relative position of the target light beam irradiated on the image acquisition unit can be changed by adjusting the optical antenna unit.

[0063] The image acquisition unit is responsible for acquiring the light spot image generated by the target light beam irradiated thereon. The light spot brightness and other information of the acquired image can be changed by changing the hardware exposure time and other parameters of the image acquisition unit. The image acquisition unit usually uses CMOS / CCD as its underlying hardware. Figure 2 The image acquisition unit is responsible for acquiring the light spot image generated by the target light beam irradiated thereon. The light spot brightness and other information of the acquired image can be changed by changing the hardware exposure time and other parameters of the image acquisition unit. The image acquisition unit usually uses CMOS / CCD as its underlying hardware.

[0064] The image processing unit reads the current acquired light spot gray value image of the image acquisition unit, and converts the acquired gray value image into a binary image by setting a gray value threshold. The obtained binary image is input into the trained convolutional neural network based on the CBAM module. The neural network outputs the image region recognition results of the main lobe region of the optical phased array, the sidelobe region of the optical phased array, and the stray light region.

[0065] The centroid recognition unit uses the centroid method to determine the centroid position of the light spot in the binary image and outputs the centroid coordinates of the main lobe region of the optical phased array far-field light spot. After converting the multi-spot centroid recognition problem into a single-spot centroid recognition problem, the centroid method can be used to determine the centroid of the light spot. Geometric method, centroid method, etc. can also be used to determine the centroid parameters according to the actual application requirements.

[0066] The negative feedback adjustment unit compares the deviation between the centroid coordinates of the main lobe region of the optical phased array far-field light spot and the center coordinates of the image acquisition unit field of view, calculates the related control parameters of the optical antenna unit, and controls the optical antenna unit to move the light spot centroid to the center of the image acquisition unit field of view.

[0067] Further, in the image processing unit region recognition strategy, the convolutional neural network target detection method based on the CBAM module is a supervised machine learning method, and its training method is as follows:

[0068] (1) Establish a data set

[0069] Collect binary images containing the main lobe of the optical phased array, the sidelobe of the optical phased array, and the stray light spot, and establish a data set. Since the pixel value in the binary image is 0 or 1, no image normalization processing is required.

[0070] (2) Image annotation

[0071] The data set is manually annotated to mark the main lobe region of the optical phased array, the sidelobe region of the optical phased array, and the stray light region. The data set of the light spot type and the light spot position is used for subsequent model training and evaluation.

[0072] (3) Model training

[0073] The CBAM module-based convolutional neural network is a multi-layer convolutional neural network, the first, third and fifth layers in the multi-layer convolutional neural network are convolutional layers, the second, fourth and sixth layers in the multi-layer convolutional neural network are CBAM layers, the seventh layer in the multi-layer convolutional neural network is a fully connected layer, and the eighth layer in the multi-layer convolutional neural network is a Softmax layer.

[0074] The CBAM (Convolutional Block Attention Module) module enhances the extraction ability of the convolutional neural network for different channel features and different position features by introducing a channel attention module and a spatial attention module. Figure 3 As shown in the figure, the channel attention module (Channel Attention Module) can enhance the feature expression of each channel; as shown in the figure, Figure 4 As shown in the figure, the spatial attention module (Spatial Attention Module) can enhance the feature expression of different positions in the image; as shown in the figure, Figure 5 As shown in the figure, the CBAM module element-wise multiplies the output features of the channel attention module and the spatial attention module to obtain the final attention-enhanced features.

[0075] In this embodiment, the 60% of the labeled binary image dataset is divided into a training set by random division, which is used to train the CBAM module-based convolutional neural network to learn data features; 20% of the dataset is divided into a validation set, which is used to adjust the hyperparameters of the model and prevent the model from overfitting; 20% of the dataset is divided into a test set, which is used to evaluate the generalization ability of the trained model.

[0076] (4) Result evaluation

[0077] In this embodiment, the mean average precision mAP (mean Average Precision) is used to evaluate the accuracy of the model prediction result, and the calculation formula is as follows:

[0078]

[0079] Where C represents the total number of categories, represents the AP value of the i-th target.

[0080] The application also discloses a CBAM-CNN-based optical phased array main lobe centroid recognition method, and the core of the method is to realize image area recognition of the main lobe area of the optical phased array, the sidelobe area of the optical phased array and the stray light area through the CBAM module-based convolutional neural network, and to convert the multi-spot recognition problem into a single-spot recognition problem in the main lobe area of the optical phased array.

[0081] In this regard, a specific embodiment is cited for specific description.

[0082] The algorithm flow of the embodiment provided by the present application is shown in Figure 6

[0083] Step 1: Set the termination condition, and end the algorithm when the number of loop cycles reaches the predetermined value.

[0084] Step 2: Adjust the optical antenna system so that the far-field spot of the optical phased array enters the field of view range of the image acquisition unit, and further adjust the optical antenna system so that the main lobe of the far-field spot of the optical phased array appears at the center of the camera field of view range.

[0085] Step 3: The default exposure time parameter of the image acquisition unit CCD / CMOS is , if the image processing unit judges that the image is overexposed, the exposure time parameter E is reduced, and the collected image is transmitted to the image processing unit.

[0086] Step 4: The image processing unit judges whether the collected image spot is overexposed, and if the image is overexposed, it jumps to Step 3.

[0087] Step 5: The image processing unit sets the image gray value threshold, and converts the collected gray value image into a binary image (pixels exceeding the current gray value threshold G are white, and pixels below the current gray value threshold are black).

[0088] Step 6: The image processing unit inputs the binary image into the convolutional neural network based on the CBAM module, and the neural network outputs the image region coordinates of the main lobe region of the optical phased array, the sidelobe region of the optical phased array, and the stray light region.

[0089] Step 7: The centroid recognition unit uses the centroid method to judge the centroid position of the spot in the main lobe region of the optical phased array, and outputs the centroid coordinates (x, y) of the main lobe region of the optical phased array far-field spot.

[0090] Step 8: Judge whether the termination condition of the algorithm is met, and if the termination condition is met, end the algorithm.

[0091] Step 9: The negative feedback adjustment unit judges whether the optical phased array far-field spot exceeds the hardware acquisition field of view of the image acquisition unit according to the centroid coordinates of the main lobe region of the optical phased array far-field spot, if it exceeds the field of view, it jumps to Step 2; if it does not exceed the field of view, it jumps to Step 4.

[0092] As a preferred, the identification of the main lobe region and the sidelobe region based on the CBAM module convolutional neural network can be realized by the following steps:

[0093] ​First, an optical phased array far-field light spot binary image dataset for training a convolutional neural network is obtained. Second, a convolutional neural network based on a CBAM module is established, the neural network is trained using the optical phased array far-field light spot binary image dataset, and a trained convolutional neural network based on a CBAM module is obtained; the first, third and fifth layers in the multi-layer convolutional neural network are convolutional layers, the second, fourth and sixth layers in the multi-layer convolutional neural network are CBAM layers, the seventh layer in the multi-layer convolutional neural network is a fully connected layer, and the eighth layer in the multi-layer convolutional neural network is a Softmax layer. The CBAM layer is composed of a channel attention module and a spatial attention module. The channel attention module has a corresponding process description formula:

[0094]

[0095] wherein, is a channel attention module weight coefficient of the intermediate feature F, is a Sigmoid function, MLP is a multi-layer neural network, AvgPool is an average pooling, and MaxPool is a maximum pooling, and is a weight of the multi-layer neural network MLP, is an average pooling value of the intermediate feature F in the channel attention module, is a maximum pooling value of the intermediate feature F in the channel attention module. The spatial attention module has a corresponding process description formula:

[0096]

[0097] wherein, is a spatial attention module weight coefficient of the intermediate feature F, is a convolution kernel size of 5x5, is an average pooling value of the intermediate feature F in the spatial attention module, is a maximum pooling value of the intermediate feature F in the spatial attention module. The CBAM layer has a corresponding process description formula:

[0098]

[0099] wherein, is an intermediate quantity of the intermediate feature F after the channel attention module, is an output quantity of the intermediate feature F after the channel attention module and the spatial attention module.

[0100] Finally, the binary image of the optical phased array far-field spot to be identified is input into the CBAM module-based convolutional neural network trained.

[0101] The above embodiments are only specific examples for further detailing the purposes, technical solutions and beneficial effects of the present application, and the present application is not limited thereto. Any modification, equivalent replacement, improvement, etc. made within the scope of the disclosure of the present application is included in the protection scope of the present application.

Claims

1. A CBAM-CNN-based optical phased array main lobe centroid recognition method, comprising the following steps: S1. Adjusting the optical antenna unit so that the optical phased array far-field spot enters the field of view range of the image acquisition unit, and adjusting the optical antenna unit so that the main lobe of the optical phased array far-field spot appears at the center of the camera field of view range; S2. Adjusting the hardware exposure parameters of the image acquisition unit so that as many spots as possible are displayed in its field of view without excessive exposure; S3. The image processing unit reads the current gray value image of the image acquisition unit, and converts the collected gray value image into a binary image by setting a gray value threshold; S4. The image processing unit inputs the obtained binary image into a CBAM module-based convolutional neural network, which outputs the main lobe region of the optical phased array, the sidelobe region of the optical phased array, and the stray light region; S5. The centroid recognition unit uses the centroid method to determine the centroid position of the main lobe region spot, and outputs the centroid coordinates of the main lobe region of the optical phased array far-field spot; S6. The negative feedback adjustment unit determines whether the optical phased array far-field spot exceeds the hardware acquisition field of view of the image acquisition unit. If it exceeds the field of view, it jumps to step S1; If it does not exceed the field of view, it jumps to step S3; In step S4, the identification of the main lobe region and the sidelobe region by the CBAM module-based convolutional neural network is achieved through the following steps: S41. Obtain an optical phased array far-field spot binary image dataset for training the convolutional neural network; S42. Establish a CBAM module-based convolutional neural network, train the neural network using the optical phased array far-field spot binary image dataset, and obtain a trained CBAM module-based convolutional neural network; S43. Input the optical phased array far-field spot binary image to be identified into the trained CBAM module-based convolutional neural network; The CBAM module-based convolutional neural network in step S42 is a multi-layer convolutional neural network, the first, third and fifth layers of the multi-layer convolutional neural network are convolutional layers, the second, fourth and sixth layers of the multi-layer convolutional neural network are CBAM layers, the seventh layer of the multi-layer convolutional neural network is a fully connected layer, and the eighth layer of the multi-layer convolutional neural network is a Softmax layer.

2. The CBAM-CNN-based optical phased array main lobe centroid identification method according to claim 1, characterized in that: The CBAM layer is composed of a channel attention module and a spatial attention module.

3. The CBAM-CNN-based optical phased array main lobe centroid identification method according to claim 2, characterized in that: The corresponding process description formula of the channel attention module is: ， wherein, is a channel attention module weight coefficient of the intermediate feature F, is a Sigmoid function, MLP is a multi-layer neural network, AvgPool is an average pooling, and MaxPool is a maximum pooling, and is a weight of the multi-layer neural network MLP, is an average pooling value of the intermediate feature F in the channel attention module, is a maximum pooling value of the intermediate feature F in the channel attention module.

4. The CBAM-CNN-based optical phased array main lobe centroid identification method according to claim 3, characterized in that: The corresponding process description formula of the spatial attention module is: ， wherein, is a spatial attention module weight coefficient of the intermediate feature F, is a convolution kernel size of 5x5, is an average pooling value of the intermediate feature F in the spatial attention module, is a maximum pooling value of the intermediate feature F in the spatial attention module.

5. The CBAM-CNN-based optical phased array main lobe centroid identification method according to claim 4, characterized in that: The corresponding process description formula of the CBAM layer is: ， wherein, is an intermediate quantity of the intermediate feature F passing through the channel attention module, is an output quantity of the intermediate feature F passing through the channel attention module and the spatial attention module.

6. The CBAM-CNN-based optical phased array main lobe centroid identification method according to claim 1, characterized in that: Steps S1 to S6 are executed in a loop, and the centroid coordinates of the main lobe region of the optical phased array far-field spot are output once every loop. The loop ends when the number of cycles reaches a predetermined value.

7. The CBAM-CNN based optical phased array main lobe centroid identification system, characterized in that, A CBAM-CNN-based optical phased array main lobe centroid recognition method as claimed in any one of claims 1-6, comprising an optical antenna unit, an image acquisition unit, an image processing unit, a centroid recognition unit, and a negative feedback adjustment unit. The optical antenna unit is used for receiving a target light beam and transmitting the light beam onto the image acquisition unit, and the relative position of the target light beam irradiated on the image acquisition unit is changed by adjusting the optical antenna unit; The image acquisition unit is used for acquiring a light spot image generated by the target light beam irradiated thereon, and the light spot brightness information of the acquired image is changed by changing the hardware exposure time parameter of the image acquisition unit; The image processing unit is used for converting the acquired light spot gray value image into a binary image, and the image region identification of the main lobe region of the optical phased array, the side lobe region of the optical phased array and the stray light region is realized by the convolutional neural network based on the CBAM module; The centroid recognition unit judges the centroid position of the light spot in the main lobe region of the optical phased array by using the centroid method, and outputs the centroid coordinates of the main lobe region of the optical phased array far-field light spot; The negative feedback adjustment unit is used for comparing the deviation of the centroid coordinates of the main lobe region of the optical phased array far-field light spot and the field center coordinates of the image acquisition unit, calculating the related control parameters of the optical antenna unit, and controlling the optical antenna unit to move the light spot centroid to the field center of the image acquisition unit.

8. The CBAM-CNN based optical phased array main lobe centroid identification system of claim 7, wherein: The image processing unit adopts a supervised machine learning method based on the CBAM module convolutional neural network target detection method when identifying the region, and the training method is as follows: (1) Establishing a data set Collect binary images containing the main lobe of the optical phased array, the side lobe of the optical phased array and the stray light spot, and establish a data set; (2) Image annotation The data set is manually annotated, and the main lobe region of the optical phased array, the side lobe region of the optical phased array, and the stray light region are marked. The data set of the light spot type and the light spot position is used for subsequent model training and evaluation; (3) Model training The CBAM module based convolutional neural network is a multi-layer convolutional neural network. The first, third and fifth layers in the multi-layer convolutional neural network are convolutional layers. The second, fourth and sixth layers in the multi-layer convolutional neural network are CBAM layers. The seventh layer in the multi-layer convolutional neural network is a fully connected layer. The eighth layer in the multi-layer convolutional neural network is a Softmax layer; The CBAM module enhances the extraction ability of the convolutional neural network for different channel features and different position features by introducing a channel attention module and a spatial attention module; (4) Result evaluation The mean average precision mAP is used to evaluate the accuracy of the model prediction result, and the calculation formula is as follows: ， where C denotes the total number of categories, represents the AP value of the i-th target category.

Images