Camera adaptive exposure processing system and method based on deep learning

Abstract

A camera adaptive exposure processing system and method based on deep learning, belonging to the field of artificial intelligence technology. In the system, a data preprocessing module preprocesses the light intensity, the distance from the camera to the subject, the input data of the special effect setting and the running speed of the carrier to convert them into data matching the input layer of the camera adaptive exposure model; the camera adaptive exposure model generates a first control value for the camera control mechanism to control the aperture size, a second control value for controlling the shutter speed and a third control value for controlling the sensitivity according to the data provided by the data preprocessing module, and the camera adaptive exposure model is trained by a neural network. The present invention can automatically generate control values ​​for controlling the aperture size, shutter speed and sensitivity according to the light intensity, the distance from the camera to the subject, the input data of the special effect setting and the speed of the camera carrier, so as to obtain photos that meet the requirements.

Description

Technical Field

[0001] The present invention relates to a camera adaptive exposure processing system and method based on deep learning, belonging to the technical field of artificial intelligence. Background Art

[0002] The Chinese invention patent application with publication number CN114900619A discloses an underwater image processing system for adaptive exposure driven camera photography, the system includes: a programmable camera and a computing and processing device; the computing and processing device includes: an adaptive exposure module and a pre-trained underwater image restoration module. The system encodes the image captured by the previous frame of the programmable camera and sends it to the adaptive exposure module. By processing the previous frame of the image, the programmable camera's shutter speed, sensitivity, aperture and other parameters are comprehensively adjusted; it is ensured that the original image captured is conducive to the deep network to extract and utilize features; finally, after the pre-trained underwater image restoration module, the quality of the intelligent body's execution of the action is judged and the color correction is output. It is ensured that the original image captured is conducive to the deep network to extract and utilize features; finally, after the pre-trained underwater image restoration module, the quality of the intelligent body's execution of the action is judged and the image after color correction and feature enhancement is output, achieving better results and improving the efficiency and accuracy of underwater operations.

[0003] However, the quality of a photo is related to the intensity of ambient light, the distance to the object being photographed, and the speed of the carrier. The above patent application does not disclose how to adjust the control parameters of the camera when taking photos according to the above influencing factors to obtain the best photo. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the object of the present invention is to provide a camera adaptive exposure processing system and method based on deep learning, which can automatically generate control values ​​for controlling the aperture size, shutter speed and sensitivity according to the intensity of the light source, the distance from the camera to the subject, the input special effects, and the camera carrier speed, so as to obtain photos that meet the requirements.

[0005] To achieve the above-mentioned purpose, the present invention also provides a camera adaptive exposure processing system based on deep learning, which includes: an ambient light intensity sensor, a distance detector, a carrier speed detector, a data preprocessing module and a camera adaptive exposure model, wherein the ambient light intensity sensor is used to obtain the light source intensity when the camera is shooting, and the distance detector is used to obtain the distance from the camera to the subject; the carrier speed detector is used to obtain the running speed of the carrier carrying the camera; the data preprocessing module preprocesses the light source intensity, the distance from the camera to the subject, the input data of the special effect setting and the running speed of the carrier to convert the light source intensity, the distance from the camera to the subject, the input data of the special effect setting and the running speed of the carrier into data matching the input layer of the camera adaptive exposure model; the camera adaptive exposure model generates a first control value for controlling the aperture size of the camera control mechanism, a second control value for controlling the shutter speed and a third control value for controlling the sensitivity according to the data provided by the data preprocessing module, and the camera adaptive exposure model is trained by a neural network.

[0006] To achieve the above-mentioned object, the present invention provides a camera adaptive exposure processing method based on deep learning, which comprises the following steps:

[0007] Step 1: The intensity of the light source when the camera is shooting is obtained through the ambient light intensity sensor, the distance from the camera to the object being photographed is obtained through the distance detector, and the running speed of the carrier carrying the camera is obtained through the carrier speed detector;

[0008] Step 2: preprocessing the light intensity, the distance from the camera to the object, the input data of the special effect setting and the running speed of the carrier through the data preprocessing module to convert the light intensity data, the distance from the camera to the object, the input data of the special effect setting and the running speed of the carrier into data matching the input layer of the camera adaptive exposure model;

[0009] Step 3: Generate a first control value for controlling the aperture size, a second control value for controlling the shutter speed, and a third control value for controlling the sensitivity of the camera control mechanism according to the data provided by the data preprocessing module through a camera adaptive exposure model, wherein the camera adaptive exposure model is trained by a neural network.

[0010] Compared with the prior art, the camera adaptive exposure processing system and method based on deep learning provided by the present invention can automatically generate control values ​​for the aperture size, shutter speed and sensitivity according to the intensity of the light source, the distance from the camera to the subject, the input special effects and the camera carrier speed, thereby obtaining photos that meet the requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a block diagram of the composition of a camera adaptive exposure processing system based on deep learning provided in the first embodiment of the present invention.

[0012] Figure 2 It is a block diagram of the composition of the signaling module of the communication unit of the camera adaptive exposure processing system based on deep learning provided in the first embodiment of the present invention.

[0013] Figure 3 It is a block diagram of the composition of the receiving module of the communication unit of the camera adaptive exposure processing system based on deep learning provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0015] In the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0016] First embodiment

[0017] Figure 1 1 is a block diagram of the composition of the signaling device of the camera adaptive exposure processing system based on deep learning provided by the first embodiment of the present invention. Figure 1 As shown, the camera adaptive exposure processing system based on deep learning provided by the first embodiment of the present invention includes: an ambient light intensity sensor, a distance detector, a carrier speed detector, a data preprocessing module and a camera adaptive exposure model, wherein the ambient light intensity sensor is used to obtain the light source intensity when the camera is shooting, and the distance detector is used to obtain the distance from the camera to the subject; the carrier speed detector is used to obtain the running speed of the carrier carrying the camera; the data preprocessing module preprocesses the light source intensity, the distance from the camera to the subject, the input data of the special effect setting and the running speed of the carrier to convert the light source intensity, the distance from the camera to the subject, the input data of the special effect setting and the running speed of the carrier into data matching the input layer of the camera adaptive exposure model; the camera adaptive exposure model generates a first control value for controlling the aperture size of the camera control mechanism, a second control value for controlling the shutter speed and a third control value for controlling the sensitivity according to the data provided by the data preprocessing module, and the camera adaptive exposure model is trained by a neural network, and the neural network is, for example, an LLM model.

[0018] In the present invention, the process of training a neural network into a camera adaptive exposure model includes:

[0019] S1-1: Obtain N photos that meet the standard in advance, and label the N photos as 1, ..., n, ..., N respectively; record the light intensity when taking the nth photo, the distance from the camera to the subject, the input data of the special effect setting, and the running speed of the carrier carrying the camera in advance, and pre-process them through the data pre-processing module to obtain pre-training data , It is the data related to the light source intensity provided by the data preprocessing module. It is the data related to the distance from the camera to the object provided by the data preprocessing module. It is the data related to the input data of the special effect settings provided by the data preprocessing module. It is the data related to the running speed of the carrier provided by the data preprocessing module; it also records in advance the aperture control value of the photographer adjusting the aperture when taking the nth photo that meets the standard , adjust the shutter speed control value of the shutter speed , adjust the sensitivity control value of the sensitivity , aperture control value , shutter speed control value and sensitivity control value Composing test data ; n=1,...,N, N is a positive integer greater than or equal to 50;

[0020] S1-2: Take out the pre-training data X of the nth photo that meets the standard n and test data ;

[0021] S1-3: pre-training data X n Input to the input layer of the neural network and output from the output layer :

[0022] ,

[0023] in, Represents the neural network fitted with parameter θ with X n is a function of the independent variable; , are the parameters of the neural network; is the predicted value of the first control value output by the neural network to control the aperture size; is a predicted value of a second control value for adjusting the shutter speed output by the neural network; is the predicted value of the third control value for adjusting the sensitivity output by the neural network;

[0024] S1-4: Calculate the first loss function according to the following formula:

[0025] ;

[0026] S1-5: Update the parameters of the neural network according to the following formula:

[0027] ;

[0028] In the formula, is the learning rate, Express The gradient;

[0029] S1-6: Pre-training data X n Weak perturbations are performed on the original data, and the weak perturbation self-supervision loss is calculated:

[0030] ,

[0031] is a hyperparameter; In pre-training data The data obtained by performing a weak perturbation on the basis of In pre-training data The data is obtained by performing two weak perturbations on the basis of Represents the neural network fitted with parameter θ is a function of the independent variable; Represents the neural network fitted with parameter θ is a function of the independent variable; T represents transpose;

[0032] S1-7: Pre-training data X n Perform strong perturbation and calculate the strong perturbation self-supervision loss:

[0033] ,

[0034] is a hyperparameter; In pre-training data The data obtained by performing a strong disturbance on the basis of In pre-training data The data obtained by performing two strong disturbances on the basis of Represents the neural network fitted with parameter θ is a function of the independent variable; Represents the neural network fitted with parameter θ is a function of the independent variable; T represents transpose;

[0035] S1-8: Update the parameters of the neural network again according to the following formula:

[0036] ;

[0037] In the formula, and is the learning rate;

[0038] S1-9: Determine whether the sum of the weak perturbation self-supervision loss and the strong perturbation self-supervision loss is the minimum. If so, end the training. If not, replace the current parameters with new parameters, assign n+1 to n, and then return to step S1-2.

[0039] The deep learning-based camera adaptive exposure processing system provided in the first embodiment of the present invention can automatically generate control values ​​for controlling the aperture size, shutter speed and sensitivity according to the light source intensity, the distance from the camera to the subject, the input special effects and the camera carrier speed through the above-mentioned technical scheme, thereby obtaining photos that meet the requirements.

[0040] The deep learning-based camera adaptive exposure processing system provided in the first embodiment also includes a communication unit. The processor transmits the photos taken by the camera to the ground terminal through the communication unit in a timely manner, and is also used to receive special effect setting data from the ground terminal.

[0041] In the first embodiment, the camera carrier is a drone.

[0042] Figure 2 is a block diagram of a signaling module of a communication unit of a camera adaptive exposure processing system based on deep learning provided by a first embodiment of the present invention, such as Figure 2 As shown, the communication unit includes a signal transmission module, the signal transmission module includes a modulation signal generation module, the modulation signal generation module includes: a first grouping module, a first code reading module, a modulation sorting module, a first iteration module, a first assignment module and a first judgment module, wherein the first grouping module groups the bit code stream to be sent, each group includes 2 M bit code; the first code reading module reads a group of 2 M bit codes; the modulation sorting module sorts the 2 M The code bits are read in reverse order to obtain 2 M bit codes are then transmitted in parallel to the first iteration module; the first iteration module is provided in parallel by the modulation sorting module according to the 2 M bit codes are iteratively operated to generate modulation symbols , and then the high-frequency modulated wave is obtained by loading it on the high-frequency signal through the modulator; the first assignment module Assign to ; The first judgment module judges Is it greater than or equal to 2? MIf no, it returns to the first iteration module and repeats the execution. If yes, it returns to the first code reading module and reads 2 M Bit code.

[0043] The first iteration module performs a second order arrangement of 2 M bit codes are iteratively run to generate modulation symbols The steps include:

[0044] S2-1: 2 will be arranged in the second order M The bit codes are grouped in pairs in order to obtain the first modulation data group, which includes 2 M-1 modulate the first group of data groups; the second data of each modulation sub-data group modulates the first complex index Multiply and then add to the first data to obtain the first complex data;

[0045] S2-2: 2 M-1 The first modulated complex data are grouped in pairs in order to obtain the second modulated data group, and the second modulated data group includes 2 M-2 modulate the second data of each modulation sub-data group of the second group of data groups and modulate the second complex index Multiply and then add to the first data to obtain modulated second complex data;

[0046] S2-3: And so on, M-m个 The mth complex data are grouped in pairs in order to obtain the mth modulated data group. The mth modulated data group includes 2 M-m modulate the mth data group of each modulation sub-data group and modulate the mth complex index The modulated data is then multiplied and then added to the first data to obtain the modulated m-th complex data, until the modulated data group M, the modulated data group M includes 1 modulated sub-data group; the modulated sub-data group 2 of the modulated data group M is modulated with the M-th complex index. Multiply and then add to the first data to get the modulation symbol .

[0047] The communication unit also includes a modulator, a high-frequency power amplifier and a transmitting antenna. The modulator modulates the information provided by the modulation signal generating module to a high frequency; the high-frequency power amplifier amplifies the power of the signal provided by the modulator, and then converts it into electromagnetic waves through the transmitting antenna and sends it into space.

[0048] Figure 3 is a block diagram of a receiving module of a communication unit of a camera adaptive exposure processing system based on deep learning provided by a first embodiment of the present invention, such as Figure 3As shown, the receiving module of the present invention includes a receiving antenna, a small signal amplifier, a demodulator and a bit code stream recovery module, wherein the receiving antenna receives electromagnetic signals from space electromagnetic waves and converts the electromagnetic signals into electrical signals, the small signal amplifier amplifies the electrical signals received by the antenna as high-frequency small signals; the demodulator extracts the modulated data sent by the transmitter from the amplified high-frequency small signals.

[0049] The bit code stream recovery module includes: a second grouping module, a second reading module, a demodulation sorting module, a second iteration module, a second assignment module and a second judgment module, wherein the second grouping module groups the received demodulated data, each group includes 2 M The second reading module reads a group of 2 modulation symbols sorted in a third order. M modulation symbols;

[0050] The demodulation sorting module sorts the 2 M Modulation symbols are read in reverse order to obtain 2 M The modulation symbols are then transmitted in parallel to the second iteration module;

[0051] The second iteration module provides the demodulation sorting module in parallel with the 2 M The modulation symbols are iteratively calculated to generate bit codes, and the second assignment module Assign to ; The second judgment module judges Is it greater than or equal to 2? M If yes, then return to the second reading module and read 2 M modulation symbols, if not, return to the second iteration module.

[0052] In the first embodiment, the second iteration module provides the 2 bits arranged in the fourth order in parallel according to the demodulation sorting module. M The iterative operation of modulation symbols to generate a bit code stream includes the following steps:

[0053] S3-1: 2 to be arranged in the fourth order M The modulation symbols are grouped in pairs in order to obtain the first demodulation data group. The first demodulation data group includes 2 M-1 demodulation sub-data group; make the second data of each demodulation sub-data group of the demodulation first group of data group and the first complex index Multiplying and then adding to the first data to obtain demodulated first complex data;

[0054] S3-2: 2 M-1 The demodulated first complex data are grouped in pairs in order to obtain a demodulated second group of data. The demodulated second group of data includes 2 M-2demodulation sub-data group; so that the second data of each demodulation sub-group data of the demodulation second group data group and the demodulation second complex index Multiplying and then adding to the second data to obtain demodulated second complex data;

[0055] S3-3: And so on, M-m The demodulated m-th complex data are grouped in order to obtain the demodulated m-th data group. The demodulated m-th data group includes 2 M-m demodulation sub-data group; the second data of each demodulation sub-group data of the demodulation m-th group data group and the demodulation m-th complex index The demodulated data is then multiplied and added to the first demodulated data to obtain the demodulated m-th complex data, until the demodulated data group M is obtained, the demodulated data group M includes a demodulated sub-data group; the demodulated second data of the demodulated sub-data group of the demodulated data group M is combined with the demodulated M complex index The bit code is multiplied and then added to the demodulated first data to obtain the bit code, which is then provided to the processor, and the processor processes the received information.

[0056] First embodiment Through the above technical solution, 2 M bit codes, and can also receive 2 M个 Bit code improves the efficiency of information transmission.

[0057] Second embodiment

[0058] The second embodiment of the present invention only describes the contents that are different from the first embodiment, and the same contents will not be described repeatedly.

[0059] A second embodiment of the present invention provides a camera adaptive exposure processing method based on deep learning, which includes the following steps:

[0060] Step 1: The intensity of the light source when the camera is shooting is obtained through the ambient light intensity sensor, the distance from the camera to the object being photographed is obtained through the distance detector, and the running speed of the carrier carrying the camera is obtained through the carrier speed detector;

[0061] Step 2: preprocessing the light intensity, the distance from the camera to the object, the input data of the special effect setting and the running speed of the carrier through the data preprocessing module to convert the light intensity data, the distance from the camera to the object, the input data of the special effect setting and the running speed of the carrier into data matching the input layer of the camera adaptive exposure model;

[0062] Step 3: Generate a first control value for controlling the aperture size, a second control value for controlling the shutter speed, and a third control value for controlling the sensitivity of the camera control mechanism according to the data provided by the data preprocessing module through a camera adaptive exposure model, wherein the camera adaptive exposure model is trained by a neural network.

[0063] The deep learning-based camera adaptive exposure processing method provided in the second embodiment of the present invention can automatically generate control values ​​for controlling the aperture size, shutter speed and sensitivity according to the light source intensity, the distance from the camera to the subject, the input special effects and the camera carrier speed through the above-mentioned technical scheme, thereby obtaining photos that meet the requirements.

[0064] The present invention also provides a computer program product, which is a program code compiled by a computer language according to the above method, and the program code can be called and executed by one or more processors.

[0065] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A camera adaptive exposure processing system based on deep learning, characterized in that: include: An ambient light intensity sensor, a distance detector, a carrier speed detector, a data preprocessing module and a camera adaptive exposure model, wherein the ambient light intensity sensor is used to obtain the light intensity when the camera is shooting, the distance detector is used to obtain the distance from the camera to the object being photographed; the carrier speed detector is used to obtain the running speed of the carrier carrying the camera; the data preprocessing module preprocesses the light intensity, the distance from the camera to the object being photographed, the input data of the special effect setting and the running speed of the carrier to convert the light intensity, the distance from the camera to the object being photographed, the input data of the special effect setting and the running speed of the carrier into data matching the input layer of the camera adaptive exposure model; the camera adaptive exposure model generates a first control value for the camera control mechanism to control the aperture size, a second control value for controlling the shutter speed and a third control value for controlling the sensitivity according to the data provided by the data preprocessing module; the camera adaptive exposure model is trained by a neural network; The process of training a neural network into a camera adaptive exposure model includes: S1-1: Obtain N photos that meet the standard in advance, and label the N photos as 1, ..., n, ..., N respectively; record the light intensity when taking the nth photo, the distance from the camera to the subject, the input data of the special effect setting, and the running speed of the carrier carrying the camera in advance, and pre-process them through the data pre-processing module to obtain pre-training data , It is the data related to the light source intensity provided by the data preprocessing module. It is the data related to the distance from the camera to the object provided by the data preprocessing module. It is the data related to the input data of the special effect settings provided by the data preprocessing module. It is the data related to the running speed of the carrier provided by the data preprocessing module; it also records in advance the aperture control value of the photographer adjusting the aperture when taking the nth photo that meets the standard , adjust the shutter speed control value of the shutter speed and the ISO control value to adjust the sensitivity , aperture control value , shutter speed control value and sensitivity control value Composing test data ; n=1,...,N, N is a positive integer greater than or equal to 50; S1-2: Take out the pre-training data X of the nth photo that meets the standard n and test data ; S1-3: pre-training data X n Input to the input layer of the neural network and output from the output layer : ， in, Represents the neural network fitted with parameter θ with X n is a function of the independent variable; , are the parameters of the neural network; is the predicted value of the first control value output by the neural network to control the aperture size; is a predicted value of a second control value for adjusting the shutter speed output by the neural network; is the predicted value of the third control value for adjusting the sensitivity output by the neural network; S1-4: Calculate the first loss function according to the following formula: ; S1-5: Update the parameters of the neural network according to the following formula: ; In the formula, is the learning rate, Express The gradient; S1-6: Pre-training data X n Weak perturbations are performed on the original data, and the weak perturbation self-supervision loss is calculated: , is a hyperparameter; In pre-training data The data obtained by performing a weak perturbation on the basis of In pre-training data The data is obtained by performing two weak perturbations on the basis of Represents the neural network fitted with parameter θ is a function of the independent variable; Represents the neural network fitted with parameter θ is a function of the independent variable; T represents transpose; S1-7: Pre-training data X n Perform strong perturbation and calculate the strong perturbation self-supervision loss: , is a hyperparameter; In pre-training data The data obtained by performing a strong disturbance on the basis of In pre-training data The data obtained by performing two strong disturbances on the basis of Represents the neural network fitted with parameter θ is a function of the independent variable; Represents the neural network fitted with parameter θ is a function of the independent variable; T represents transpose; S1-8: Update the parameters of the neural network again according to the following formula: ; In the formula, and is the learning rate; S1-9: Determine whether the sum of the weak perturbation self-supervision loss and the strong perturbation self-supervision loss is the minimum. If so, end the training. If not, replace the current parameters with new parameters, assign n+1 to n, and then return to step S1-2.

2. The camera adaptive exposure processing system based on deep learning according to claim 1, characterized in that: The camera carrier is a drone.

3. A camera adaptive exposure processing method based on deep learning, characterized in that: The steps include: Step 1: The intensity of the light source when the camera is shooting is obtained through the ambient light intensity sensor, the distance from the camera to the object being photographed is obtained through the distance detector, and the running speed of the carrier carrying the camera is obtained through the carrier speed detector; Step 2: preprocessing the light intensity, the distance from the camera to the object, the input data of the special effect setting and the running speed of the carrier through the data preprocessing module to convert the light intensity data, the distance from the camera to the object, the input data of the special effect setting and the running speed of the carrier into data matching the input layer of the camera adaptive exposure model; Step 3: Generate a first control value for controlling the aperture size, a second control value for controlling the shutter speed, and a third control value for controlling the sensitivity of the camera control mechanism according to the data provided by the data preprocessing module through a camera adaptive exposure model, wherein the camera adaptive exposure model is trained by a neural network; The process of training a neural network into a camera adaptive exposure model includes: S1-1: Obtain N photos that meet the standard in advance, and label the N photos as 1, ..., n, ..., N respectively; record the light intensity when taking the nth photo, the distance from the camera to the subject, the input data of the special effect setting, and the running speed of the carrier carrying the camera in advance, and pre-process them through the data pre-processing module to obtain pre-training data , It is the data related to the light source intensity provided by the data preprocessing module. It is the data related to the distance from the camera to the object provided by the data preprocessing module. It is the data related to the input data of the special effect settings provided by the data preprocessing module. It is the data related to the running speed of the carrier provided by the data preprocessing module; it also records in advance the aperture control value of the photographer adjusting the aperture when taking the nth photo that meets the standard , adjust the shutter speed control value of the shutter speed , adjust the sensitivity control value of the sensitivity , aperture control value , shutter speed control value and sensitivity control value Composing test data ; n=1,...,N, N is a positive integer greater than or equal to 50; S1-2: Take out the pre-training data X of the nth photo that meets the standard n and test data ; S1-3: pre-training data X n Input to the input layer of the neural network and output from the output layer : ， in, Represents the neural network fitted with parameter θ with X n is a function of the independent variable; , are the parameters of the neural network; is the predicted value of the first control value output by the neural network to control the aperture size; is a predicted value of a second control value for adjusting the shutter speed output by the neural network; is the predicted value of the third control value for adjusting the sensitivity output by the neural network; S1-4: Calculate the first loss function according to the following formula: ; S1-5: Update the parameters of the neural network according to the following formula: ; In the formula, is the learning rate, Express The gradient; S1-6: Pre-training data X n Weak perturbations are performed on the original data, and the weak perturbation self-supervision loss is calculated: , is a hyperparameter; In pre-training data The data obtained by performing a weak perturbation on the basis of In pre-training data The data is obtained by performing two weak perturbations on the basis of Represents the neural network fitted with parameter θ is the function value of the independent variable; Represents the neural network fitted with parameter θ is the function value of the independent variable; T represents transposition; S1-7: Pre-training data X n Perform strong perturbation and calculate the strong perturbation self-supervision loss: , is a hyperparameter; In pre-training data The data obtained by performing a strong disturbance on the basis of In pre-training data The data obtained by performing two strong disturbances on the basis of Represents the neural network fitted with parameter θ is the function value of the independent variable; Represents the neural network fitted with parameter θ is the function value of the independent variable; T represents transposition; S1-8: Update the parameters of the neural network again according to the following formula: ; In the formula, and is the learning rate; S1-9: Determine whether the sum of the weak perturbation self-supervision loss and the strong perturbation self-supervision loss is the minimum. If so, end the training. If not, replace the current parameters with new parameters, assign n+1 to n, and then return to step S1-2.

4. The camera adaptive exposure processing method based on deep learning according to claim 3, characterized in that: The camera carrier is a drone.

Images