OCR recognition system based on big data analysis

Abstract

The invention discloses an OCR recognition system based on big data analysis, and relates to the technical field of optical character recognition, and the system comprises an environment data collection module, a data preprocessing module, a big data analysis and processing module, a model selection and adjustment module, an OCR recognition module and a feedback optimization module. The environmental adaptability of the OCR recognition system is improved by applying various sensors and a complex data preprocessing technology, the system can collect environmental humidity, temperature, light, vibration and inclination multi-dimensional data in real time, the accuracy and reliability of the data are improved through the steps of noise removal and normalization processing of a filter, and the accuracy and reliability of the OCR recognition system are improved. In the aspect of preprocessing of character image data, a character area can be automatically cut, a background part can be automatically removed, and adaptive image enhancement processing including histogram equalization under low illumination and gamma correction under strong light can be carried out according to light intensity, so that the influence of environmental factors on an OCR recognition effect is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of optical character recognition, and in particular to an OCR recognition system based on big data analysis. Background Art

[0002] With the rapid development of information technology, OCR (Optical Character Recognition) technology, as an efficient and automated text recognition method, has been widely used in document scanning, image processing, and automated data entry, greatly improving the efficiency of data processing and information extraction. OCR technology can convert text in images into editable and searchable text formats, providing great convenience for users.

[0003] Traditional OCR technology mainly relies on static image processing and pattern recognition algorithms, and lacks sufficient adaptability and robustness to changes in environmental factors. For example, when humidity, temperature, and light environmental conditions change significantly, traditional OCR technology may experience problems such as increased recognition error rate and decreased recognition speed. In addition, traditional OCR technology has relatively limited processing capabilities for noise and deformation interference factors in images, which further limits its application scope in complex environments. At the same time, traditional OCR technology often uses a single recognition model, which is difficult to flexibly adjust according to changes in environmental complexity, thus affecting its recognition effect.

[0004] Therefore, the development of OCR recognition system based on big data analysis will open up new paths for the development and application of OCR technology. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the prior art and provide an OCR recognition system based on big data analysis. By integrating multiple sensors and complex data processing technologies, it realizes the real-time collection and accurate processing of multi-dimensional data such as environmental humidity, temperature, light, vibration and tilt, and can intelligently select the most appropriate OCR recognition model according to the complexity of the environment. In addition, the present invention also continuously optimizes the system performance and recognition model by introducing feedback optimization to ensure that the system can continue to adapt to various complex environments.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an OCR recognition system based on big data analysis, the system comprising: an environmental data acquisition module, a data preprocessing module, a big data analysis and processing module, a model selection and adjustment module, an OCR recognition module and a feedback optimization module;

[0007] Environmental data acquisition module: By arranging capacitive humidity sensors, thermocouple temperature sensors, light sensors including photodiodes and optical filters, piezoelectric vibration sensors, and MEMS three-axis tilt sensors around the OCR device, it can obtain environmental humidity, temperature, light, vibration, and tilt data in real time, and simultaneously obtain the text image data to be recognized;

[0008] Data preprocessing module: The various environmental data collected by the sensor are normalized after noise removal using a Butterworth low-pass filter. For text image data, the text area is automatically cropped and irrelevant background parts are removed. The image is enhanced by an adaptive method based on the light intensity, including histogram equalization in low light and gamma correction in strong light. At the same time, median filtering is used to remove salt and pepper noise and Gaussian noise in the image.

[0009] Big data analysis and processing module: Receive pre-processed sensor data, use Hadoop HDFS to store environment and recognition data, and use MySQL to manage data metadata. The environmental complexity comprehensive evaluation formula is used to quantify the complexity of the environment for OCR recognition. Let E be the environmental complexity, H be the humidity value, T be the temperature value, L be the light intensity, V be the vibration intensity, I be the tilt angle, ω1, ω2, ω3, ω 4、 ω5 is the corresponding weight, and the calculation formula is: Among them, H0, T0, and L0 are the ideal reference values ​​of humidity, temperature, and light intensity, respectively. max , H min 、T max , L max , L min are the maximum and minimum values ​​of the measurement range of humidity, temperature, and light intensity, respectively. The degree of deformation and blurring of text caused by humidity and temperature changes is predicted by the text deformation and blurring degree prediction formula. Let D HT is the degree of text deformation and blurring caused by humidity and temperature, α and β are the coefficients of humidity and temperature on text deformation, and the calculation formula is: D HT =α×(H-H0) 2 +β×(T-T0) 2 ;

[0010] Model selection and adjustment module: Establish a model library containing multiple OCR models, and the comprehensive evaluation value of environmental complexity E is lower than the set threshold E th , the model selection and adjustment module selects a simple OCR model, and the comprehensive evaluation value of environmental complexity E exceeds the threshold E th , select a complex model based on the type and degree of environmental change;

[0011] OCR recognition module: In text recognition, the corresponding algorithms are used to perform recognition, and the environmental information and text deformation prediction and correction results are combined, and the recognition parameters are adjusted according to the characteristics of different text carriers such as paper and display screens;

[0012] Feedback optimization module: Feedback the OCR recognition results and intermediate data to the big data analysis module, evaluate the system performance through evaluation indicators, and make optimization suggestions to the big data analysis and model adjustment modules after comparing historical results and finding problems.

[0013] Furthermore, the data preprocessing module performs histogram equalization under low light conditions, calculates the grayscale histogram of the image, and assumes that the grayscale range of the image I(x, y) is [0, L-1], where x and y are the pixel coordinates of the image, and counts the frequency p(k) of each grayscale level k (0≤k≤L-1), and the calculation formula is: Among them, n k is the number of pixels with gray level k, M×N is the total number of pixels in the image, and the cumulative distribution function CDF is calculated as follows: Calculate the grayscale value T(k) after equalization using the formula: T(k)=round((L-1)×C(k)), where the round function rounds the result to the nearest integer. Replace the grayscale value of the pixel with the grayscale value k in the original image with T(k) to obtain the image after histogram equalization.

[0014] Furthermore, in the gamma correction under strong light in the data preprocessing module, assuming that the grayscale value of the original image pixel is I(x, y), the grayscale value after gamma correction is O(x, y), the gamma value is γ and γ>0, then the gamma correction formula is:

[0015] Furthermore, the gamma value in the data preprocessing module is calculated based on the average brightness B of the image. avg To adjust the gamma value γ, the calculation formula is: γ=1+k×(B avg -B th ), where k is the proportionality factor, B th is the strong light judgment threshold, the average brightness B avg , let the pixel coordinates of the image be (x, y), the brightness value be I(x, y), the total number of pixels in the image be M×N, and the average brightness calculation formula be: The strong light judgment threshold is B th =350cd / m 2 .

[0016] Furthermore, the environment complexity threshold E in the model selection and adjustment module thDetermine, calculate the comprehensive evaluation formula value E of environmental complexity through the comprehensive evaluation formula of environmental complexity, record the recognition accuracy A of the simple OCR recognition model in each environment, and the calculation formula is: Where N correct is the number of correctly recognized characters, N total is the total number of characters to be recognized. Draw a curve of the relationship between the environment complexity E and the recognition accuracy A, observe the changing trend of the curve, and take the point of environment complexity where the recognition accuracy drops significantly as the threshold E th .

[0017] Furthermore, the complex OCR recognition model based on convolutional neural network in the model selection and adjustment module: a CNN architecture with multiple convolutional layers, pooling layers and fully connected layers is designed. A simple CNN architecture includes: a convolutional layer, using multiple convolutional kernels to perform convolution operations on the input text image, assuming that the convolution kernel size is k×k, the step size is s, and the number of convolution kernels is n. For the input image I, the feature map F after the convolution operation is calculated as: Where ω is the convolution kernel weight, b is the bias term, i, j are the coordinates of the feature map, pooling layer: add maximum pooling after the convolution layer to reduce the data dimension, set the pooling window size to p×p, the step size to s, for the feature map F, the calculation formula of the feature map P after pooling is: Fully connected layer: After several convolution and pooling layers, the feature map is flattened into a one-dimensional vector and connected to one or more fully connected layers. The number of neurons in the fully connected layer gradually decreases, and finally the number of nodes in the output layer is equal to the number of text categories to be recognized.

[0018] Furthermore, the calculation formula of the simple OCR model in the model selection and adjustment module is: Where P(y|x) is the probability that the text belongs to category y given the input image feature x, n is the number of training samples, and x is i is the feature vector of the i-th training sample, α i is the parameter related to the i-th training sample, K(x,x i ) is a custom kernel function, and the calculation formula is: K(x,x i )=(a(x·x i )+r) d , where a is a tuning parameter used to control the scaling of the kernel function, r is a constant, and d is the degree of the polynomial used to measure the difference between the input feature x and the training sample feature x i The similarity between them, b is the bias term, and C is the total number of text categories.

[0019] Furthermore, in the OCR recognition module, the image feature vector is adjusted by the image feature adjustment formula under the influence of the light spectrum when the light spectrum changes. Let F L is the adjusted image feature vector, F0 is the original image feature vector, b is the light spectrum influence coefficient, which is obtained by analyzing the image features under different light spectra, S is the light spectrum feature vector, and the calculation formula is: F L =F0×(1+b×S).

[0020] Furthermore, the OCR recognition module corrects the image deformation caused by vibration and tilt through the image correction formula under the influence of vibration and tilt. corrected is the corrected image, I is the original image, δ V is the vibration influence function, δ I is the tilt influence function, which is obtained by modeling the relationship between vibration and tilt data and image changes. The calculation formula is: corrected =I×(1-δ V (V))×(1-δ I (I)).

[0021] Compared with the existing technology, the OCR recognition system based on big data analysis has the following beneficial effects:

[0022] 1. The present invention improves the environmental adaptability of the OCR recognition system by using a variety of sensors and complex data preprocessing technologies. The system can collect multi-dimensional data such as environmental humidity, temperature, light, vibration and tilt in real time, and remove noise and normalize the data through filters to improve the accuracy and reliability of the data. In the preprocessing of text image data, it can automatically crop the text area, remove the background part, and perform adaptive image enhancement processing according to the light intensity, including histogram equalization in low light and gamma correction in strong light, thereby reducing the impact of environmental factors on the OCR recognition effect and improving the accuracy and stability of recognition.

[0023] 2. The present invention realizes the intelligent selection and adjustment of OCR recognition models by constructing a model library containing multiple OCR models and introducing a comprehensive evaluation mechanism for environmental complexity. It can calculate the comprehensive evaluation value of environmental complexity based on the environmental data collected in real time, and select the most appropriate OCR recognition model for text recognition accordingly. This mechanism not only improves the recognition efficiency, but also can flexibly adjust the model for environments of different complexities, thereby improving the overall performance and adaptability of the OCR recognition system.

[0024] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 This is a flowchart of the OCR recognition system based on big data analysis;

[0027] Figure 2 Flowchart of the big data analysis and processing module. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. 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.

[0029] Embodiment 1:

[0030] Digitalization of ancient books in libraries

[0031] Data collection, various sensors are arranged around the OCR equipment used for scanning ancient books. The capacitive humidity sensor detects that the indoor humidity of the library is H, the thermocouple temperature sensor shows the temperature is T, the light sensor (including photodiode and optical filter) measures the light intensity as L, the piezoelectric vibration sensor and MEMS three-axis tilt sensor do not detect obvious vibration, and at the same time obtain the text image data to be recognized on the ancient book page.

[0032] Data preprocessing: The environmental data collected by the sensor is normalized after the noise is removed by a Butterworth low-pass filter. For the ancient text image data, the text area is automatically cropped and the surrounding irrelevant background is removed. Due to the low light intensity, the image is enhanced by histogram equalization. Assuming that the grayscale range of the image I (x, y) is [0, L-1], where x and y are the pixel coordinates of the image, the frequency p (k) of each grayscale level k (0≤k≤L-1) is counted, and the calculation formula is: Among them, n kis the number of pixels with gray level k, M×N is the total number of pixels in the image, and the cumulative distribution function CDF is calculated as follows: Calculate the gray value T(k) after equalization. The formula is: T(k)=round((L-1)×C(k)), where the round function rounds the result to the nearest integer. Replace the gray value of the pixel with the gray value k in the original image with T(k) to obtain the image after histogram equalization. At the same time, use median filtering to remove a small amount of salt and pepper noise and Gaussian noise in the image.

[0033] Big data analysis and processing, pre-processed sensor data is input, Hadoop HDFS is used to store the environment and identify data, and MySQL is used to manage data metadata. Assume that H is the humidity value, T is the temperature value, L is the light intensity, V is the vibration intensity, I is the tilt angle, ω1, ω2, ω3, ω 4、 ω5 is the corresponding weight. The environmental complexity E is calculated according to the comprehensive evaluation formula of environmental complexity. The calculation formula is: Then, the degree of text deformation and blurring caused by humidity and temperature changes is predicted by the text deformation and blurring prediction formula. The calculation formula is: HT =α×(H-H0) 2 +β×(T-T0) 2 , where α and β are the coefficients of humidity and temperature on text deformation.

[0034] Model selection and adjustment, the calculated environment complexity E and the set threshold E th Compare the environment complexity threshold E th Determine, calculate the comprehensive evaluation formula value E of environmental complexity through the comprehensive evaluation formula of environmental complexity, record the recognition accuracy A of the simple OCR recognition model in each environment, and the calculation formula is: Where N correct is the number of correctly recognized characters, N total is the total number of characters to be recognized. Draw a curve of the relationship between the environment complexity E and the recognition accuracy A, observe the changing trend of the curve, and take the point of environment complexity where the recognition accuracy drops significantly as the threshold E th If E is lower than E th , select a simple OCR recognition model that uses manual features combined with machine learning, and the comprehensive evaluation value of environmental complexity E is lower than the set threshold E th , the model selection and adjustment module selects a simple OCR model calculation formula: Where P(y|x) is the probability that the text belongs to category y given the input image feature x, n is the number of training samples, and x is i is the feature vector of the i-th training sample, α iis the parameter related to the i-th training sample, K(x,x i ) is a custom kernel function, and the calculation formula is: K(x,x i )=(a(x·x i )+r) d , where a is a tuning parameter used to control the scaling of the kernel function, r is a constant, and d is the degree of the polynomial used to measure the difference between the input feature x and the training sample feature x i The similarity between them, b is the bias term, and C is the total number of text categories.

[0035] In the OCR recognition process, the corresponding algorithm is used when performing recognition. The change of sunlight angle in different time periods causes the light spectrum on the book to change. The image feature vector is adjusted when the light spectrum changes through the image feature adjustment formula under the influence of the light spectrum. Let F L is the adjusted image feature vector, F0 is the original image feature vector, b is the light spectrum influence coefficient, which is obtained by analyzing the image features under different light spectra, S is the light spectrum feature vector, and the calculation formula is: F L =F0×(1+b×S), for the image deformation caused by vibration and tilt, the image correction formula under the influence of vibration and tilt is used to correct the image deformation caused by vibration and tilt. Let I corrected is the corrected image, I is the original image, δ V is the vibration influence function, δ I is the tilt influence function, which is obtained by modeling the relationship between vibration and tilt data and image changes. The calculation formula is: corrected =I×(1-δ V (V))×(1-δ I (I)) Adjust the recognition parameters according to the characteristics of ancient paper carriers, such as adjusting the character segmentation and recognition threshold according to the paper color and texture.

[0036] Feedback optimization: Feedback the OCR recognition results and intermediate data to the big data analysis module, evaluate the system performance through evaluation indicators, compare historical results, find some text recognition errors or reduced accuracy, and make optimization suggestions to the big data analysis and model adjustment modules, such as adjusting model parameters and improving image preprocessing methods.

[0037] In summary, in the scenario of ancient book digitization in the library, this embodiment arranges multiple sensors around the OCR device to collect environmental data and ancient book text image data, and uses the big data analysis and processing module to calculate the environmental complexity E and the text deformation blur degree D HT , based on E and the set threshold E thBy comparing and selecting the appropriate OCR recognition model, the environmental information and text deformation prediction and correction results are combined in the recognition process, and the parameters are adjusted according to the characteristics of the paper carriers of ancient books. Finally, after feedback optimization, the system performance is evaluated based on the recognition results and intermediate data, and optimization suggestions are put forward. This effectively improves the accuracy and adaptability of ancient book text recognition, provides reliable technical support for the digitization of ancient books, and helps to protect and inherit cultural heritage.

[0038] Embodiment 2:

[0039] Outdoor billboard text recognition

[0040] Data collection, various sensors are installed on the OCR device used to shoot outdoor billboards. The capacitive humidity sensor detects the outdoor humidity as H, the thermocouple temperature sensor displays the temperature as T, the light sensor measures the light intensity as L, the piezoelectric vibration sensor detects the vehicle vibration (vibration intensity V), and the MEMS three-axis tilt sensor detects that the device is tilted due to hand-held or vehicle driving (tilt angle I), and at the same time obtains the text image data to be recognized on the billboard.

[0041] Data preprocessing: Environmental data is denoised and normalized by a Butterworth low-pass filter. For text image data, the text area is cropped and the background is removed. Due to the strong light, gamma correction is performed under strong light and the gamma value is adjusted according to the average brightness of the image. Suppose the pixel coordinates of the image are (x, y), the brightness value is I(x, y), the total number of pixels of the image is M×N, and the average brightness is B avg The calculation formula is: Correct the bright words and use median filtering to remove image noise.

[0042] Big data analysis and processing, pre-processed data is input, and Hadoop HDFS and MySQL are used to store and manage data. Assume that E is the environmental complexity, H is the humidity value, T is the temperature value, L is the light intensity, V is the vibration intensity, I is the tilt angle, ω1, ω2, ω3, ω 4、 ω5 is the corresponding weight. According to the comprehensive evaluation formula of environmental complexity, the calculation formula is: At the same time, the text deformation and blurring degree prediction formula is used to predict the degree of text deformation and blurring caused by humidity and temperature changes. Let D HT D is the degree of text deformation and blurring caused by humidity and temperature, α and β are the coefficients of humidity and temperature on text deformation, and the calculation formula is: HT =α×(H-H0) 2 +β×(T-T0) 2 .

[0043] Model selection and adjustment, based on the calculated environment complexity E and threshold E th Compare (the threshold determination method is the same as in Example 1), if E exceeds E th , choose a complex OCR recognition model, such as a model based on a convolutional neural network (CNN). The designed CNN architecture includes convolutional layers (convolution kernel size k×k, step size s, number of convolution kernels n parameters. For the input image I, the feature map F after the convolution operation is calculated as: Where ω is the convolution kernel weight, b is the bias term, i, j are the coordinates of the feature map), pooling layer (pooling window size p×p, step size s, for feature map F, the feature map after pooling And fully connected layer: flatten the feature map into a one-dimensional vector, connect it to one or more fully connected layers, the number of neurons in the fully connected layer gradually decreases, and finally the number of output layer nodes is equal to the number of text categories to be recognized.

[0044] The OCR recognition process relies on the convolutional neural network to automatically extract text features. The corresponding algorithm is used when performing recognition to adjust the image feature vector when the light spectrum changes. For example, the change in the angle of sunlight in different time periods causes the light spectrum of the billboard to change. The image feature vector is adjusted using the image feature adjustment formula under the influence of the light spectrum. The calculation formula is: F L =F0×(1+b×S), where the light spectrum influence coefficient b is obtained by analyzing the image characteristics under different light spectra, and the light spectrum feature vector S is determined according to the actual light spectrum. For image deformation caused by vibration and tilt, the image correction formula under the influence of vibration and tilt is used for correction. The calculation formula is: I corrected =I×(1-δ V (V))×(1-δ I (I)), vibration influence function δ V (V) and the tilt influence function δ I (I) Obtained by modeling the relationship between vibration and tilt data and image changes.

[0045] Feedback optimization: The recognition results and intermediate data are fed back to evaluate the system performance. If it is found that the recognition accuracy is reduced due to complex outdoor environmental factors (such as strong light reflection and blurred text caused by aging billboards), optimization suggestions are made to the big data analysis and model adjustment modules, such as adding image preprocessing methods for strong light environments and adjusting CNN model parameters to adapt to the characteristics of outdoor text.

[0046] In summary, in the outdoor billboard text recognition embodiment, the sensor installed in the OCR device obtains the environment and billboard text image data, and the environment data is denoised and normalized during data preprocessing. The gamma value is adjusted according to the average brightness of the image under strong light to correct the image. E and D are calculated through big data analysis and processing.HT , according to E and E th The relationship selection model is used. During recognition, the complex model automatically extracts features with the help of convolutional neural networks, and the feature vector and the corrected image are adjusted based on the light spectrum change and vibration tilt influence formula. Finally, feedback optimization is used to improve system performance. This embodiment adapts to complex outdoor environments, ensures the effectiveness and stability of billboard text recognition, and has important application value in the fields of urban management and commercial advertising monitoring.

[0047] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. The OCR recognition system based on big data analysis is characterized by: The system includes: environmental data acquisition module, data preprocessing module, big data analysis and processing module, model selection and adjustment module, OCR recognition module and feedback optimization module; Environmental data acquisition module: By arranging capacitive humidity sensors, thermocouple temperature sensors, light sensors including photodiodes and optical filters, piezoelectric vibration sensors, and MEMS three-axis tilt sensors around the OCR device, it can obtain environmental humidity, temperature, light, vibration, and tilt data in real time, and simultaneously obtain the text image data to be recognized; Data preprocessing module: The various environmental data collected by the sensor are normalized after noise removal using a Butterworth low-pass filter. For text image data, the text area is automatically cropped and irrelevant background parts are removed. The image is enhanced by an adaptive method based on the light intensity, including histogram equalization in low light and gamma correction in strong light. At the same time, median filtering is used to remove salt and pepper noise and Gaussian noise in the image. Big data analysis and processing module: Receive pre-processed sensor data, use Hadoop HDFS to store environment and recognition data, and use MySQL to manage data metadata. The environmental complexity comprehensive evaluation formula is used to quantify the complexity of the environment for OCR recognition. Let E be the environmental complexity, H be the humidity value, T be the temperature value, L be the light intensity, V be the vibration intensity, I be the tilt angle, ω1, ω2, ω3, ω 4、 ω5 is the corresponding weight, and the calculation formula is: Among them, H0, T0, and L0 are the ideal reference values ​​of humidity, temperature, and light intensity, respectively. max , H min , T max , L max , L min are the maximum and minimum values ​​of the measurement range of humidity, temperature, and light intensity, respectively. The degree of deformation and blurring of text caused by humidity and temperature changes is predicted by the text deformation and blurring degree prediction formula. Let D HT D is the degree of text deformation and blurring caused by humidity and temperature, α and β are the coefficients of humidity and temperature on text deformation, and the calculation formula is: HT =α×(H-H0) 2 +β×(T-T0) 2 ; Model selection and adjustment module: Establish a model library containing multiple OCR models, and the comprehensive evaluation value of environmental complexity E is lower than the set threshold E th , the model selection and adjustment module selects a simple OCR model, and the comprehensive evaluation value of environmental complexity E exceeds the threshold E th , select a complex model based on the type and degree of environmental change; OCR recognition module: In text recognition, the corresponding algorithms are used to perform recognition, and the environmental information and text deformation prediction and correction results are combined, and the recognition parameters are adjusted according to the characteristics of different text carriers such as paper and display screens; Feedback optimization module: Feedback the OCR recognition results and intermediate data to the big data analysis module, evaluate the system performance through evaluation indicators, and make optimization suggestions to the big data analysis and model adjustment modules after comparing historical results and finding problems.

2. The OCR recognition system based on big data analysis according to claim 1, characterized in that: The data preprocessing module performs histogram equalization under low light conditions and calculates the grayscale histogram of the image. Assuming that the grayscale range of the image I(x, y) is [0, L-1], where x and y are the pixel coordinates of the image, the frequency p(k) of each grayscale level k (0≤k≤L-1) is counted, and the calculation formula is: Among them, n k is the number of pixels with gray level k, M×N is the total number of pixels in the image, and the cumulative distribution function CDF is calculated as follows: Calculate the grayscale value T(k) after equalization using the formula: T(k)=round((L-1)×C(k)), where the round function rounds the result to the nearest integer. Replace the grayscale value of the pixel with the grayscale value k in the original image with T(k) to obtain the image after histogram equalization.

3. The OCR recognition system based on big data analysis according to claim 1, characterized in that: In the data preprocessing module, for gamma correction under strong light, the gray value of the original image pixel is I(x, y), the gray value after gamma correction is O(x, y), the gamma value is γ and γ>0, then the gamma correction formula is:

4. The OCR recognition system based on big data analysis according to claim 3 is characterized in that: The gamma value in the data preprocessing module is based on the average brightness B of the image. avg To adjust the gamma value γ, the calculation formula is: γ=1+k×(B avg -B th ), where k is the proportionality factor, B th is the strong light judgment threshold, the average brightness B avg , let the pixel coordinates of the image be (x, y), the brightness value be I(x, y), the total number of pixels in the image be M×N, and the average brightness calculation formula be: The strong light judgment threshold is B th =350cd / m 2 .

5. The OCR recognition system based on big data analysis according to claim 1, characterized in that: The environment complexity threshold E in the model selection and adjustment module th Determine, calculate the comprehensive evaluation formula value E of environmental complexity through the comprehensive evaluation formula of environmental complexity, record the recognition accuracy A of the simple OCR recognition model in each environment, and the calculation formula is: Where N correct is the number of correctly recognized characters, N total is the total number of characters to be recognized. Draw a curve of the relationship between the environment complexity E and the recognition accuracy A, observe the changing trend of the curve, and take the point of environment complexity where the recognition accuracy drops significantly as the threshold E th .

6. The OCR recognition system based on big data analysis according to claim 1, characterized in that: The complex OCR recognition model based on convolutional neural network in the model selection and adjustment module: Design a CNN architecture with multiple convolutional layers, pooling layers and fully connected layers. A simple CNN architecture includes: a convolutional layer, using multiple convolutional kernels to perform convolution operations on the input text image. Assuming the convolution kernel size is k×k, the step size is s, and the number of convolution kernels is n, for the input image I, the feature map F after the convolution operation is calculated as: Where ω is the convolution kernel weight, b is the bias term, i, j are the coordinates of the feature map, pooling layer: add maximum pooling after the convolution layer to reduce the data dimension, set the pooling window size to p×p, the step size to s, for the feature map F, the calculation formula of the feature map P after pooling is: Fully connected layer: After several convolution and pooling layers, the feature map is flattened into a one-dimensional vector and connected to one or more fully connected layers. The number of neurons in the fully connected layer gradually decreases, and finally the number of nodes in the output layer is equal to the number of text categories to be recognized.

7. The OCR recognition system based on big data analysis according to claim 1, characterized in that: The calculation formula of the simple OCR model in the model selection and adjustment module is: Where P(y|x) is the probability that the text belongs to category y given the input image feature x, n is the number of training samples, and x is i is the feature vector of the i-th training sample, α i is the parameter related to the i-th training sample, K(x,x i ) is a custom kernel function, and the calculation formula is: K(x,x i )=(a(x·x i )+r) d , where a is a tuning parameter used to control the scaling of the kernel function, r is a constant, and d is the degree of the polynomial used to measure the difference between the input feature x and the training sample feature x i The similarity between them, b is the bias term, and C is the total number of text categories.

8. The OCR recognition system based on big data analysis according to claim 1, characterized in that: In the OCR recognition module, the image feature vector is adjusted by the image feature adjustment formula under the influence of the light spectrum when the light spectrum changes. Let F L is the adjusted image feature vector, F0 is the original image feature vector, b is the light spectrum influence coefficient, which is obtained by analyzing the image features under different light spectra, S is the light spectrum feature vector, and the calculation formula is: F L =F0×(1+b×S).

9. The OCR recognition system based on big data analysis according to claim 1, characterized in that: The OCR recognition module corrects the image deformation caused by vibration and tilt through the image correction formula under the influence of vibration and tilt. corrected is the corrected image, I is the original image, δ V is the vibration influence function, δ I is the tilt influence function, which is obtained by modeling the relationship between vibration and tilt data and image changes. The calculation formula is: corrected =I×(1-δ V (V))×(1-δ I (I)).

Images