Financial statement automatic processing method based on statistical analysis and image recognition

Abstract

The invention relates to a financial statement automatic processing method based on statistical analysis and image recognition, and belongs to the field of target image recognition. The method comprises the following steps: obtaining a financial statement sample image, and obtaining a financial statement image data set; an improved YOLOv8-based financial statement target recognition network model is constructed, the model comprises an Input input end, a Backbone network, a Neck network and a Head output end, and financial statement image recognition is performed on images in the financial statement image data set through the model to obtain financial statement image text information; the method comprises the following steps: carrying out pandas data analysis on financial statement image text information through a time sequence analysis method, processing a data analysis result by using matplotlib and a seaborn library, generating a related trend chart, and realizing visualization; and based on a data analysis result and the generated related trend chart, creating a financial report in a Word document format through a python-docx library. According to the method, the time for manually inputting the data can be shortened, and the working efficiency is improved.

Description

Technical Field

[0001] The invention belongs to the field of target image recognition, and in particular relates to an automatic processing method for financial statements based on statistical analysis and image recognition. Background Art

[0002] Financial statements are an important reflection of a company's financial status and operating results. Their processing and analysis are of great significance to company management and investment decisions. Traditional financial statement processing methods mainly rely on manual operations, which are inefficient, error-prone, costly, and difficult to standardize, resulting in uneven quality of financial statements. With the increase in the amount of corporate data, traditional manual financial statement processing methods can no longer meet the needs of modern enterprises. Manual data entry is prone to errors and has a slow processing speed, which seriously affects the timeliness of financial decisions. In recent years, image recognition and statistical analysis technologies have made significant progress in many fields, but their application in automated processing of financial statements is still limited.

[0003] At present, image recognition technology has been widely used in many industries, especially in document processing, automated data entry and intelligent auditing. Through deep learning models, such as the YOLO series, computers can quickly and accurately identify target objects in images. The application of this technology has greatly improved the speed and accuracy of data extraction and reduced human intervention. However, the existing algorithms are limited by the accuracy of image recognition algorithms and it is difficult to accurately identify key information in financial statements. The existing technology often requires manual data cleaning and format conversion, which is inefficient. The existing technology mainly focuses on information extraction and lacks statistical analysis functions for financial data.

[0004] In addition, image recognition alone cannot meet the complete financial data processing needs. It also needs to be combined with statistical analysis to effectively interpret and utilize data. Statistical analysis methods such as regression analysis, trend analysis, and descriptive statistics have shown important value in data-driven decision-making. However, existing statistical analysis tools often lack the ability to intelligently process data sources, which makes it difficult for companies to obtain accurate financial insights in a fast-changing market environment. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a method for automatic processing of financial statements based on statistical analysis and image recognition.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions: In a first aspect, the present invention provides a method for automated processing of financial statements based on statistical analysis and image recognition, comprising the following steps: S1. Obtain a sample image of a financial statement to obtain a financial statement image dataset; S2. Construct a financial statement target recognition network model based on improved YOLOv8, wherein the financial statement target recognition network model based on improved YOLOv8 includes: an input end, a backbone network, a neck network, and a head output end. The images in the financial statement image data set are recognized by the financial statement target recognition network model based on improved YOLOv8 to obtain financial statement image text information; S3. Perform pandas data analysis on the text information of financial statements through time series analysis method to obtain data analysis results. Use matplotlib and seaborn libraries to generate relevant trend charts from the data analysis results to achieve visualization; S4. Based on the results of data analysis and the generated trend charts, the financial report in Word document format is created using the python-docx library.

[0007] Furthermore, step S1 specifically includes: Set up an automated script, based on which the sample data of the financial statements of publicly available Chinese listed companies are regularly obtained, and redundant information in the data is removed to obtain a data set of financial statement images. .

[0008] Furthermore, the operation of the Input terminal in step S2 is specifically as follows: The Input input end includes a Mosaic data enhancement module, an adaptive anchor frame module and an adaptive image scaling module; Financial Statement Image Dataset The image in the input terminal is used to obtain the preprocessed financial statement image dataset , , Indicates A sample of preprocessed financial statement images.

[0009] Furthermore, the operations of the Backbone network in step S2 are specifically as follows: The Backbone network includes a convolution Conv module, a C2 feature extraction layer C2f_1, a feature extraction layer C2f_Star module and an AIFI module; the convolution Conv module includes Conv_1, Conv_2, Conv_3, Conv_4, Conv_5 and Conv6, and the feature extraction layer C2f_Star module includes C2f_Star_1, C2f_Star_2 and C2f_Star_3; the C2f_Star_1 includes a convolution Conv1 module, 2 Star Blocks modules and a convolution Conv2 module; the Star Blocks module includes a depth-separable convolution DW-Conv1, 3 1×1 fully connected layers and a depth-separable convolution DW-Conv2, and the structures of C2f_Star_2 and C2f_Star_3 are the same as those of C2f_Star_1; Image samples from the preprocessed financial statement image dataset After the convolution layer Conv_1, the first feature map with 64 output channels is obtained. , the first feature map After the convolution layer Conv_2, the second feature map with 128 output channels is obtained. , the second feature map After the feature extraction layer C2f_1, the third feature map is obtained , the third feature map After the convolution layer Conv_3, the fourth feature map with 256 output channels is , the fourth characteristic map The fifth feature map with 256 output channels after the feature extraction layer C2f_Star_1 , the fifth characteristic map After the convolution layer Conv_4, the sixth feature map with 512 output channels is , the sixth characteristic graph After the feature extraction layer C2f_Star_2, the seventh feature map with an output channel number of 512 is obtained , the seventh characteristic diagram After the convolution layer Conv_5, the output channel number is 1024, the eighth feature map , the eighth characteristic map After the feature extraction layer C2f_Star_3, the ninth feature map with an output channel number of 1024 is obtained , the ninth characteristic graph After downsampling in the convolutional layer Conv_6, the number of channels becomes 512, and then it is input into the AIFI module to perform in-scale interaction on the high-level semantic features, and the tenth feature map with an output channel number of 512 is obtained. .

[0010] Furthermore, the operation of the Neck network in step S2 is specifically as follows: The Neck network includes a feature-focused diffusion pyramid network FCF module, a convolutional layer Conv module, a C2 feature extraction layer, a Concat connection layer and an upsampled layer Upsample module; the feature-focused diffusion pyramid network FCF module includes FCF_1 and FCF_2, the convolutional layer Conv module includes Conv_6, Conv_7 and Conv_8, the C2 feature extraction layer includes C2f_2, C2f_3, C2f_4 and C2f_5, the Concat connection layer includes Concat_1, Concat_2, Concat_3 and Concat_4, and the upsampled layer Upsample module includes Upsample_1 and Upsample_2; The seventh characteristic diagram 、The tenth characteristic graph With the fifth characteristic diagram After the feature focused diffusion pyramid network FCF_1, the eleventh feature map with an output channel number of 128 is obtained , the eleventh feature map After the convolution layer Conv_6, the twelfth feature map with an output channel number of 256 is obtained , the twelfth characteristic graph With the tenth characteristic graph After the Concat connection layer Concat_1 is connected, the thirteenth feature map with an output channel number of 256 is obtained , the thirteenth characteristic diagram After the feature extraction layer C2f_2, the fourteenth feature map with an output channel number of 256 is obtained , the eleventh characteristic graph After the upsampling layer Upsample module Upsample_1, the upsampling is doubled, and the output is the fifteenth feature map with 256 output channels. , the fifteenth characteristic graph With the fifth characteristic diagram Through the Concat connection layer Concat_2 splicing, the sixteenth feature map with 256 output channels is obtained , the sixteenth characteristic diagram After the feature extraction layer C2f_3, the seventeenth feature map with 256 output channels is obtained. , the seventeenth characteristic diagram , Eleventh characteristic graph With the fourteenth characteristic figure After the feature focused diffusion pyramid network FCF_2, the eighteenth feature map with 256 output channels is obtained. , the eighteenth characteristic diagram After the Conv_7 module, the nineteenth feature map with an output channel number of 512 is obtained , the nineteenth characteristic diagram With the twelfth characteristic diagram 、The fourteenth characteristic graph Through the Concat_3 module, the twentieth feature map with an output channel number of 1024 is obtained , the twentieth feature map After the feature extraction layer C2f_4, the twenty-first feature map with an output channel number of 1024 is obtained , the twenty-first feature map After the upsampling layer Upsample module Upsample_2, the twenty-second feature map is obtained , the fifteenth characteristic diagram 、22nd characteristic graph With the seventeenth characteristic figure After the Concat connection layer Concat_3, the twenty-third feature map with an output channel number of 1024 is obtained , the twenty-third feature map After the feature extraction layer C2f_5, the twenty-fourth feature map with an output channel number of 1024 is obtained .

[0011] Furthermore, the operation of the Head output terminal in step S2 is specifically as follows: The Head output end is composed of a Detect_DyHead module; the Detect_DyHead module includes a convolutional layer Conv module and a DyHeadBlock module; The eighteenth characteristic diagram 、The twenty-fourth characteristic diagram The detection frame is generated by the Detect_Dyhead detection module, the recognition of the financial statement image is completed, and the text information of the financial statement image is obtained; The financial statement image text information includes financial indicators, time stamps, market data and macroeconomic data.

[0012] Furthermore, step S3 specifically includes: The data analysis results include the changing trends of financial indicators over time, cyclical fluctuations in the data, the correlation between different financial indicators and the correlation between macroeconomic indicators; the relevant trend charts include monthly income comparison chart, monthly target completion rate line chart, monthly completion rate percentage chart, monthly income comparison table, current ratio table, debt ratio table and profit rate table.

[0013] In a second aspect, a financial statement automatic processing device based on statistical analysis and image recognition comprises: Data acquisition unit: used to acquire sample images of financial statements and obtain a financial statement image data set; Model building unit: used to build a financial statement target recognition network model based on improved YOLOv8, wherein the financial statement target recognition network model based on improved YOLOv8 includes: an input end, a backbone network, a neck network, and a head output end; images in the financial statement image data set are recognized by the financial statement target recognition network model based on improved YOLOv8 to obtain financial statement image text information; Data analysis unit: used to perform pandas data analysis on the text information of financial statements through time series analysis method, obtain data analysis results, and generate relevant trend charts from the data analysis results by using matplotlib and seaborn libraries to achieve visualization; Report generation unit: used to create financial reports in Word document format based on the results of data analysis and generated trend charts through the python-docx library.

[0014] In a third aspect, an electronic device includes: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the above-mentioned automated processing method of financial statements based on statistical analysis and image recognition is performed.

[0015] In a fourth aspect, a computer-readable storage medium stores a computer program, which, when executed by a processor, executes the above-mentioned method for automated processing of financial statements based on statistical analysis and image recognition.

[0016] The advantages of the present invention are: The method of the present invention greatly reduces the time of manual data input and improves work efficiency. In addition, statistical analysis and image recognition technology can accurately identify and input data, significantly reduce the error rate, and can quickly process a large amount of financial data. At the same time, it reduces labor costs, which is a huge advantage for companies with a large number of financial statements. Fast, accurate and efficient financial statement processing can improve the financial management level of the company and enhance the overall competitiveness of the company. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 The recognition accuracy of different algorithms of the present invention; Figure 3 A chart generated after data analysis of the present invention; Figure 4 Financial statements generated for the method of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Example 1 In this embodiment, Figure 1 As shown, the present invention provides a method for automatic processing of financial statements based on statistical analysis and image recognition, and the specific steps include: S1. Obtain a sample image of a financial statement to obtain a financial statement image dataset; Specifically, an automated script is set up, and based on the automated script, sample data of financial statements of publicly available Chinese listed companies is obtained on a regular basis, and redundant information in the data is removed to obtain a financial statement image dataset. .

[0021] S2. Construct a financial statement target recognition network model based on improved YOLOv8, wherein the financial statement target recognition network model based on improved YOLOv8 includes: an input end, a backbone network, a neck network, and a head output end. The images in the financial statement image data set are recognized by the financial statement target recognition network model based on improved YOLOv8 to obtain financial statement image text information; Specifically, the Input input end includes a Mosaic data enhancement module, an adaptive anchor frame module, and an adaptive image scaling module; a financial statement image dataset The image in the input terminal is used to obtain the preprocessed financial statement image dataset , , Indicates A sample of preprocessed financial statement images.

[0022] Specifically, the Backbone network includes a convolution Conv module, a C2 feature extraction layer C2f_1, a feature extraction layer C2f_Star module and an AIFI module; the convolution Conv module includes Conv_1, Conv_2, Conv_3, Conv_4, Conv_5, and Conv6, and the feature extraction layer C2f_Star module includes C2f_Star_1, C2f_Star_2, and C2f_Star_3; the C2f_Star_1 includes 1 convolution Conv1 module, 2 Star Blocks modules, and 1 convolution Conv2 module; the Star Blocks module includes 1 depth-separable convolution DW-Conv1, 3 1×1 fully connected layers, and 1 depth-separable convolution DW-Conv2, and the structures of C2f_Star_2 and C2f_Star_3 are the same as those of C2f_Star_1; Image samples from the preprocessed financial statement image dataset After the convolution layer Conv_1, the first feature map with 64 output channels is obtained. , the first feature map After the convolution layer Conv_2, the second feature map with 128 output channels is obtained. , the second feature map After the feature extraction layer C2f_1, the third feature map is obtained , the third feature map After the convolution layer Conv_3, the fourth feature map with 256 output channels is , the fourth characteristic map The fifth feature map with 256 output channels after the feature extraction layer C2f_Star_1 , the fifth characteristic map After the convolution layer Conv_4, the sixth feature map with 512 output channels is , the sixth characteristic graph After the feature extraction layer C2f_Star_2, the seventh feature map with an output channel number of 512 is obtained , the seventh characteristic diagram After the convolution layer Conv_5, the output channel number is 1024, the eighth feature map , the eighth characteristic map After the feature extraction layer C2f_Star_3, the ninth feature map with an output channel number of 1024 is obtained , the ninth characteristic graph After downsampling in the convolutional layer Conv_6, the number of channels becomes 512, and then it is input into the AIFI module to perform in-scale interaction on the high-level semantic features, and the tenth feature map with an output channel number of 512 is obtained. .

[0023] Specifically, the Neck network includes a feature-focused diffusion pyramid network FCF module, a convolutional layer Conv module, a C2 feature extraction layer, a Concat connection layer and an upsampling layer Upsample module; the feature-focused diffusion pyramid network FCF module includes FCF_1 and FCF_2, the convolutional layer Conv module includes Conv_6, Conv_7 and Conv_8, the C2 feature extraction layer includes C2f_2, C2f_3, C2f_4 and C2f_5, the Concat connection layer includes Concat_1, Concat_2, Concat_3 and Concat_4, and the upsampling layer Upsample module includes Upsample_1 and Upsample_2; The seventh characteristic diagram 、The tenth characteristic graph With the fifth characteristic diagram After the feature focused diffusion pyramid network FCF_1, the eleventh feature map with an output channel number of 128 is obtained , the eleventh feature map After the convolution layer Conv_6, the twelfth feature map with an output channel number of 256 is obtained , the twelfth characteristic graph With the tenth characteristic graph After the Concat connection layer Concat_1 is connected, the thirteenth feature map with an output channel number of 256 is obtained , the thirteenth characteristic diagram After the feature extraction layer C2f_2, the fourteenth feature map with an output channel number of 256 is obtained , the eleventh characteristic graph After the upsampling layer Upsample module Upsample_1, the upsampling is doubled, and the output is the fifteenth feature map with 256 output channels. , the fifteenth characteristic graph With the fifth characteristic diagram Through the Concat connection layer Concat_2 splicing, the sixteenth feature map with 256 output channels is obtained , the sixteenth characteristic diagram After the feature extraction layer C2f_3, the seventeenth feature map with 256 output channels is obtained. , the seventeenth characteristic diagram , Eleventh characteristic graph With the fourteenth characteristic figure After the feature focused diffusion pyramid network FCF_2, the eighteenth feature map with 256 output channels is obtained. , the eighteenth characteristic diagram After the Conv_7 module, the nineteenth feature map with an output channel number of 512 is obtained , the nineteenth characteristic diagram With the twelfth characteristic diagram 、The fourteenth characteristic graph Through the Concat_3 module, the twentieth feature map with an output channel number of 1024 is obtained , the twentieth feature map After the feature extraction layer C2f_4, the twenty-first feature map with an output channel number of 1024 is obtained , the twenty-first feature map After the upsampling layer Upsample module Upsample_2, the twenty-second feature map is obtained , the fifteenth characteristic diagram 、22nd characteristic graph With the seventeenth characteristic figure After the Concat connection layer Concat_3, the twenty-third feature map with an output channel number of 1024 is obtained , the twenty-third feature map After the feature extraction layer C2f_5, the twenty-fourth feature map with an output channel number of 1024 is obtained .

[0024] Specifically, the Head output end is composed of a Detect_DyHead module; the Detect_DyHead module includes a convolutional layer Conv module and a DyHeadBlock module; The eighteenth characteristic diagram 、The twenty-fourth characteristic diagram The detection frame is generated by the Detect_Dyhead detection module, the recognition of the financial statement image is completed, and the text information of the financial statement image is obtained; The financial statement image text information includes financial indicators, time stamps, market data and macroeconomic data.

[0025] S3. Perform pandas data analysis on the text information of financial statements through time series analysis method to obtain data analysis results. Use matplotlib and seaborn libraries to generate relevant trend charts from the data analysis results to achieve visualization; Specifically, the data analysis results include the changing trends of financial indicators over time, cyclical fluctuations in the data, the correlation between different financial indicators, and the correlation between macroeconomic indicators; the relevant trend charts include monthly income comparison charts, monthly target completion rate line charts, monthly completion rate percentage charts, monthly income comparison tables, current ratio tables, debt ratio tables, and profit rate tables.

[0026] S4. Based on the results of data analysis and the generated trend charts, the financial report in Word document format is created using the python-docx library. Example 2 In this embodiment, Table 1 is the experimental comparison data of the financial statement target recognition network model based on improved YOLOv8 of the present invention and other detection algorithms. It can be seen from Table 1 that the recognition rate and accuracy of the proposed model are higher than those of the existing model, which can better improve work efficiency, adapt to the financial management needs of various enterprises, and have wide practicality and promotion value.

[0027] Table 1 Experimental comparison data of the method of the present invention and other detection algorithms In this embodiment, Figure 2 The recognition accuracy of the method of the present invention and other detection algorithms at a certain time point is compared. Figure 2 It can be seen that the improved model has the highest recognition accuracy, which is higher than the existing model. It is fast and can obtain updated financial information in real time, which greatly improves the efficiency of report preparation and updating.

[0028] from Figure 3 As can be seen in the figure, the present invention uses the improved model to intuitively display the annual financial data in the form of visual charts, aiming to improve the comprehension and operability of financial statements and facilitate financial analysis and management decision-making. The improved model combines a variety of data presentation methods, including bar charts, line charts, pie charts and tables, to comprehensively display core financial indicators such as revenue targets, actual revenues, and target completion rates for different months.

[0029] Figure 4 For the generated financial statements, the improved model can analyze financial data from multiple angles, supporting management to conduct in-depth financial analysis and decision-making from a multi-dimensional perspective. Through the comprehensive presentation of various charts, managers can quickly identify financial health, discover potential problems and formulate improvement measures.

[0030] Example 3 This embodiment provides a financial statement automation processing device based on statistical analysis and image recognition, including: Data acquisition unit: used to acquire sample images of financial statements and obtain a financial statement image data set; Model building unit: used to build a financial statement target recognition network model based on improved YOLOv8, wherein the financial statement target recognition network model based on improved YOLOv8 includes: an input end, a backbone network, a neck network, and a head output end; images in the financial statement image data set are recognized by the financial statement target recognition network model based on improved YOLOv8 to obtain financial statement image text information; Data analysis unit: used to perform pandas data analysis on the text information of financial statements through time series analysis method, obtain data analysis results, and generate relevant trend charts from the data analysis results by using matplotlib and seaborn libraries to achieve visualization; Report generation unit: used to create financial reports in Word document format based on the results of data analysis and generated trend charts through the python-docx library.

[0031] Example 4 This embodiment provides an electronic device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the above-mentioned method for automatic processing of financial statements based on statistical analysis and image recognition is executed; The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0032] Example 5 This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned method for automatically processing financial statements based on statistical analysis and image recognition is executed; Storage media include: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, and other media that can store programs.

[0033] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for automated processing of financial statements based on statistical analysis and image recognition, characterized in that: The following steps are involved: S1. Obtain a sample image of a financial statement to obtain a financial statement image dataset; S2. Construct a financial statement target recognition network model based on improved YOLOv8, wherein the financial statement target recognition network model based on improved YOLOv8 includes: an input end, a backbone network, a neck network, and a head output end. The images in the financial statement image data set are recognized by the financial statement target recognition network model based on improved YOLOv8 to obtain financial statement image text information; S3. Perform pandas data analysis on the text information of financial statements through time series analysis method to obtain data analysis results. Use matplotlib and seaborn libraries to generate relevant trend charts from the data analysis results to achieve visualization; S4. Based on the results of data analysis and the generated trend charts, the financial report in Word document format is created using the python-docx library.

2. The method for automated processing of financial statements based on statistical analysis and image recognition according to claim 1, characterized in that: The step S1 specifically includes: Set up an automated script, based on which the sample data of the financial statements of publicly available Chinese listed companies are regularly obtained, and redundant information in the data is removed to obtain a data set of financial statement images. .

3. The method for automated processing of financial statements based on statistical analysis and image recognition according to claim 2, characterized in that: The operation of the Input terminal in step S2 is specifically as follows: The Input input end includes a Mosaic data enhancement module, an adaptive anchor frame module and an adaptive image scaling module; Financial Statement Image Dataset The image in the input terminal is used to obtain the preprocessed financial statement image dataset , , Indicates A sample of preprocessed financial statement images.

4. The method for automated processing of financial statements based on statistical analysis and image recognition according to claim 3, characterized in that: The operations of the Backbone network in step S2 are as follows: The Backbone network includes a convolution Conv module, a C2 feature extraction layer C2f_1, a feature extraction layer C2f_Star module and an AIFI module; the convolution Conv module includes Conv_1, Conv_2, Conv_3, Conv_4, Conv_5 and Conv6, and the feature extraction layer C2f_Star module includes C2f_Star_1, C2f_Star_2 and C2f_Star_3; the C2f_Star_1 includes a convolution Conv1 module, 2 Star Blocks modules and a convolution Conv2 module; the Star Blocks module includes a depth-separable convolution DW-Conv1, 3 1×1 fully connected layers and a depth-separable convolution DW-Conv2, and the structures of C2f_Star_2 and C2f_Star_3 are the same as those of C2f_Star_1; Image samples from the preprocessed financial statement image dataset After the convolution layer Conv_1, the first feature map with 64 output channels is obtained. , the first feature map After the convolution layer Conv_2, the second feature map with 128 output channels is obtained. , the second feature map After the feature extraction layer C2f_1, the third feature map is obtained , the third feature map After the convolution layer Conv_3, the fourth feature map with 256 output channels is , the fourth characteristic map The fifth feature map with 256 output channels after the feature extraction layer C2f_Star_1 , the fifth characteristic map After the convolution layer Conv_4, the sixth feature map with 512 output channels is , the sixth characteristic map After the feature extraction layer C2f_Star_2, the seventh feature map with an output channel number of 512 is obtained , the seventh characteristic diagram After the convolution layer Conv_5, the output channel number is 1024, the eighth feature map , the eighth characteristic map After the feature extraction layer C2f_Star_3, the ninth feature map with an output channel number of 1024 is obtained , the ninth characteristic graph After downsampling in the convolutional layer Conv_6, the number of channels becomes 512, and then it is input into the AIFI module to perform in-scale interaction on the high-level semantic features, and the tenth feature map with an output channel number of 512 is obtained. .

5. The method for automated processing of financial statements based on statistical analysis and image recognition according to claim 4, characterized in that: The specific operation of the Neck network in step S2 is: The Neck network includes a feature-focused diffusion pyramid network FCF module, a convolutional layer Conv module, a C2 feature extraction layer, a Concat connection layer and an upsampled layer Upsample module; the feature-focused diffusion pyramid network FCF module includes FCF_1 and FCF_2, the convolutional layer Conv module includes Conv_6, Conv_7 and Conv_8, the C2 feature extraction layer includes C2f_2, C2f_3, C2f_4 and C2f_5, the Concat connection layer includes Concat_1, Concat_2, Concat_3 and Concat_4, and the upsampled layer Upsample module includes Upsample_1 and Upsample_2; The seventh characteristic diagram , the tenth characteristic graph With the fifth characteristic diagram After the feature focused diffusion pyramid network FCF_1, the eleventh feature map with an output channel number of 128 is obtained , the eleventh feature map After the convolution layer Conv_6, the twelfth feature map with an output channel number of 256 is obtained , the twelfth characteristic graph With the tenth characteristic graph After the Concat connection layer Concat_1 is connected, the thirteenth feature map with an output channel number of 256 is obtained , the thirteenth characteristic diagram After the feature extraction layer C2f_2, the fourteenth feature map with an output channel number of 256 is obtained , the eleventh characteristic graph After the upsampling layer Upsample module Upsample_1, the upsampling is doubled, and the output is the fifteenth feature map with 256 output channels. , the fifteenth characteristic graph With the fifth characteristic diagram Through the Concat connection layer Concat_2 splicing, the sixteenth feature map with 256 output channels is obtained , the sixteenth characteristic diagram After the feature extraction layer C2f_3, the seventeenth feature map with 256 output channels is obtained. , the seventeenth characteristic diagram , Eleventh characteristic graph With the fourteenth characteristic figure After the feature focused diffusion pyramid network FCF_2, the eighteenth feature map with 256 output channels is obtained. , the eighteenth characteristic diagram After the Conv_7 module, the nineteenth feature map with an output channel number of 512 is obtained , the nineteenth characteristic diagram With the twelfth characteristic diagram 、The fourteenth characteristic graph Through the Concat_3 module, the twentieth feature map with an output channel number of 1024 is obtained , the twentieth feature map After the feature extraction layer C2f_4, the twenty-first feature map with an output channel number of 1024 is obtained , the twenty-first feature map After the upsampling layer Upsample module Upsample_2, the twenty-second feature map is obtained , the fifteenth characteristic diagram 、22nd characteristic graph With the seventeenth characteristic figure After the Concat connection layer Concat_3, the twenty-third feature map with an output channel number of 1024 is obtained , the twenty-third feature map After the feature extraction layer C2f_5, the twenty-fourth feature map with an output channel number of 1024 is obtained .

6. The method for automated processing of financial statements based on statistical analysis and image recognition according to claim 5, characterized in that: The operation of the Head output terminal in step S2 is specifically as follows: The Head output end is composed of a Detect_DyHead module; the Detect_DyHead module includes a convolutional layer Conv module and a DyHeadBlock module; The eighteenth characteristic diagram 、The twenty-fourth characteristic diagram The detection frame is generated by the Detect_Dyhead detection module, the recognition of the financial statement image is completed, and the text information of the financial statement image is obtained; The financial statement image text information includes financial indicators, time stamps, market data and macroeconomic data.

7. The method for automated processing of financial statements based on statistical analysis and image recognition according to claim 1, characterized in that: Step S3 specifically includes: The data analysis results include the changing trends of financial indicators over time, cyclical fluctuations in the data, the correlation between different financial indicators and the correlation between macroeconomic indicators; the relevant trend charts include monthly income comparison chart, monthly target completion rate line chart, monthly completion rate percentage chart, monthly income comparison table, current ratio table, debt ratio table and profit rate table.

8. An automatic processing device for financial statements based on statistical analysis and image recognition, characterized in that: include: Data acquisition unit: used to acquire sample images of financial statements and obtain a financial statement image data set; Model building unit: used to build a financial statement target recognition network model based on improved YOLOv8, wherein the financial statement target recognition network model based on improved YOLOv8 includes: an input end, a backbone network, a neck network, and a head output end; images in the financial statement image data set are recognized by the financial statement target recognition network model based on improved YOLOv8 to obtain financial statement image text information; Data analysis unit: used to perform pandas data analysis on the text information of financial statements through time series analysis method, obtain data analysis results, and generate relevant trend charts from the data analysis results by using matplotlib and seaborn libraries to achieve visualization; Report generation unit: used to create financial reports in Word document format based on the results of data analysis and generated trend charts through the python-docx library.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the method for automated processing of financial statements based on statistical analysis and image recognition as described in any one of claims 1 to 7 is performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the automated processing method for financial statements based on statistical analysis and image recognition as described in any one of claims 1 to 7.

Images