Method, system, equipment and medium for improving data volume by modifying data quality
Through machine learning, deep learning and reinforcement learning algorithms, data resolution modification is solved, and sensors from different manufacturers of the same type have poor algorithm generalization capabilities, and data quality improvement and quantity expansion are achieved, supporting seamless transplantation of algorithm models and 0 mark training.
Patent Information
- Application Number
- CN202411264251.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2024-12-13
AI Technical Summary
In the fields of robotics and autonomous driving, sensors from different manufacturers of the same type have poor generalization capabilities, and they need to re-acquire, label and train data, which cannot achieve seamless transplantation of algorithms.
Data resolution modification is carried out through machine learning, deep learning and reinforcement learning algorithms, downsampling and upsampling process original data, improve data quality and quantity, and enhance the algorithm model's generalization ability of different sensor hardware.
The generalization ability of algorithm models on different sensors is improved, reducing the need for re-acquisition and labeling, supporting 0-labeling training, and improving the richness and consistency of data samples.
Smart Images
Figure CN119130833A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technologies such as robots and autonomous driving, and more specifically, to a method, system, device and medium for increasing data volume by modifying data quality. Background Art
[0002] In fields that are highly dependent on artificial intelligence technology, such as robotics and autonomous driving, the effectiveness of algorithm reasoning directly determines the performance and safety of the system. In order to achieve efficient and accurate algorithm reasoning, high-quality sample data is an indispensable cornerstone. Data labeling is a delicate and time-consuming task that requires manual review of each piece of data and accurate labeling (such as pedestrians, vehicles, road signs, etc.). Although open source data can alleviate the problem of data scarcity to a certain extent, its quality and consistency are often difficult to guarantee. Differences between different data sources, inconsistent labeling standards, data noise and bias, etc. may affect the generalization ability of the training model and the actual deployment effect.
[0003] In the existing environment, there are similar solutions by adding mosaics, color enhancement, etc. When the resolution and imaging capabilities of sensors vary greatly, algorithms trained with other data sets are prone to perform poorly on the current sensors, that is, the generalization ability of the hardware is poor, resulting in the need to re-label and re-train algorithms deployed on sensors of the same type but from different manufacturers.
[0004] The hardware differences between sensors of the same type (referring to cameras and radar sensors) from different manufacturers will lead to large differences in the quality of image and radar data. This is not only reflected in the size of the data, such as the difference in image resolution of 1920*1080 and 640*360, but also in the different quality caused by the color differences between different pixel blocks of the sensor's photoreceptors. For example, the color difference between each two color blocks of some sensors is very obvious, and they can capture the true color of the surrounding environment very well. However, for some sensors, the difference between different color blocks is relatively small, and they cannot fully capture the true color of the surrounding environment. There is usually a certain degree of distortion, which will lead to poor generalization ability of the algorithm model for different sensors. When encountering different sensors, it is usually necessary to re-collect, label and train data. Summary of the invention
[0005] An object of the present invention is to provide a method, system, device and medium for increasing data quantity by modifying data quality, so as to solve the problem that the aforementioned algorithm model has poor generalization ability on sensors of the same type but from different manufacturers, and cannot achieve seamless algorithm transplantation, requiring re-data collection, labeling and training.
[0006] According to a first aspect of the present invention, there is provided a method for increasing data quantity by modifying data quality, comprising the following steps:
[0007] Data acquisition step: acquiring raw data from one or more data sources through a preset data acquisition device; the raw data includes but is not limited to images, videos, audio, text, and sensor signals;
[0008] Resolution modification step: using one or more combinations of machine learning models, deep learning networks, and reinforcement learning algorithms to perform resolution modification processing on the original data obtained in the data acquisition step;
[0009] Result acquisition step: After the resolution modification process is completed, the data after the resolution modification is obtained and saved.
[0010] Optionally, the method for increasing data volume by modifying data quality also includes: optimization and verification steps: performing quality assessment and optimization processing on the acquired data after the modified resolution, including but not limited to denoising, enhancement and feature extraction, and at the same time, evaluating the effect of the data after the modified resolution in actual application through a preset verification mechanism.
[0011] Optionally, the machine learning model includes but is not limited to decision trees, random forests, support vector machines, neural networks; the deep learning network includes but is not limited to convolutional neural networks (CNN), generative adversarial networks (GAN), recurrent neural networks (RNN) and their variants.
[0012] Optionally, the processing configuration in the resolution modification step is dynamically adjustable, and the processing parameters are adjusted or the processing algorithm is switched in real time according to the requirements of the actual application scenario or user instructions.
[0013] Optionally, the data acquisition device is a camera, the acquired raw data is image data, and the resolution of the raw image data is configured as a first resolution;
[0014] Downsampling the original image data to downsample the image resolution to a second resolution, where the second resolution is lower than the first resolution;
[0015] Use bilinear interpolation to upsample and restore the image resolution to the first resolution;
[0016] Add the restored image data to the data sample.
[0017] Optionally, downsampling is performed by using the average of the nearest pixels of a block with a length and a width of three pixels.
[0018] Optionally, the data acquisition device is a radar, and the acquired raw data is raw radar frame data, and the raw radar frame data is configured with a first line number;
[0019] By deleting data of a part of fixed line bundles in the original radar frame data, the original radar frame data is modified to a second line number, where the second line number is lower than the first line number;
[0020] Add the modified radar frame data to the data sample.
[0021] According to a second aspect of the present invention, there is provided a system for increasing data quantity by modifying data quality, comprising:
[0022] Data acquisition module: acquires raw data from one or more data sources through a preset data acquisition device; the raw data includes but is not limited to images, videos, audio, text, and sensor signals;
[0023] Resolution modification module: using one or more combinations of machine learning models, deep learning networks, and reinforcement learning algorithms to perform resolution modification on the original data obtained in the data acquisition step;
[0024] Result acquisition module: After the resolution modification process is completed, the data after the resolution modification is obtained and saved.
[0025] According to a third aspect of the present invention, there is provided an electronic device, the electronic device comprising: a processor and a memory;
[0026] The memory stores instructions executable by the processor;
[0027] When the processor is configured to execute the instructions, the electronic device implements the above-mentioned method of increasing data volume by modifying data quality.
[0028] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium, the computer-readable storage medium comprising: computer software instructions;
[0029] When the computer software instructions are executed in an electronic device, the electronic device implements the above-mentioned method of increasing data quantity by modifying data quality.
[0030] The method, system, device and medium for improving data volume by modifying data quality according to the present disclosure have the following technical effects:
[0031] The present invention realizes the resolution modification of the original data through algorithms such as machine learning, deep learning, and reinforcement learning. For example, the image resolution is downsampled to a lower resolution by using the average of the nearest pixels of a block with a length and width of three pixels, and then upsampled by using bilinear interpolation to restore the image, and the restored low-quality data is added to the data sample, thereby increasing the data volume. The generalization ability of the algorithm model to different sensor hardware can be improved; under certain circumstances, algorithm training with zero annotation can be realized (using open source data sets completely).
[0032] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
[0034] Figure 1 A schematic diagram of a method for increasing data volume by modifying data quality provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0035] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.
[0036] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.
[0037] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.
[0038] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0039] In the process of improving sample richness by adding mosaics, color enhancement, etc., when the resolution and imaging capabilities of sensors vary greatly, algorithms trained with other data sets are prone to perform poorly on the current sensor, that is, poor generalization capabilities on hardware, resulting in algorithms deployed on sensors of the same type but from different manufacturers needing to be re-labeled and retrained. This application proposes an algorithmic idea of expanding data by modifying data quality, and implements resolution modification of original data through machine learning, deep learning, reinforcement learning or traditional algorithms. When processing data samples, the sample size can be directly expanded to improve the generalization capabilities of trained algorithms on different hardware.
[0040] The present invention proposes an embodiment of a method for improving data quantity by modifying data quality. Specifically, Figure 1 As shown, the following steps are included:
[0041] Data acquisition step: obtain raw data from one or more data sources through a preset data acquisition device; the raw data includes but is not limited to images, videos, audio, text, and sensor signals; wherein the data acquisition device includes a camera or radar, a camera such as a digital SLR camera, a high-definition webcam, an infrared camera, or a depth camera (such as an RGB-D camera), etc.; when the data acquisition task is relatively simple or specific, only one data source (such as a single camera or radar system) needs to be acquired for data; in complex application scenarios, data needs to be collected from multiple data sources at the same time, such as combining multiple cameras for 360-degree panoramic monitoring, or combining camera and radar data for target tracking and identification.
[0042] Resolution modification step: Use one or more combinations of machine learning models, deep learning networks, and reinforcement learning algorithms to modify the resolution of the original data obtained in the data acquisition step; specifically, for simple resolution modification tasks, such as fixed-ratio upsampling or downsampling, machine learning algorithms, such as regression-based or interpolation-based methods, can be used. Deep learning, especially convolutional neural networks (CNNs), can generate more visually realistic high-resolution images by learning the mapping relationship between low-resolution images and high-resolution images for resolution modification, especially super-resolution (SR) tasks. The hyperparameters in the deep learning model can be automatically adjusted through reinforcement learning algorithms to obtain the best resolution modification effect.
[0043] Result acquisition step: After the resolution modification process is completed, the data after the resolution modification is obtained and saved.
[0044] In an embodiment of the present invention, the method for increasing the amount of data by modifying the data quality also includes an optimization and verification step: performing quality assessment and optimization processing on the acquired data after the modified resolution, including but not limited to denoising, enhancement and feature extraction, and at the same time, evaluating the effect of the data after the modified resolution in practical applications through a preset verification mechanism. Specifically, preset evaluation indicators such as peak signal-to-noise ratio (PSNR), structural similarity (SSIM), mean square error (MSE), etc. are used to quantify the image quality. These indicators can be automatically calculated to provide an objective quality score, and further check whether the modified data meets the expected resolution requirements and remains consistent throughout the data set. In addition, denoising algorithms such as Gaussian filtering, median filtering, bilateral filtering, etc. are applied to reduce noise in images or videos and improve the signal-to-noise ratio. The image is enhanced, such as brightness adjustment, contrast enhancement, color correction, etc., to improve the visual effect, and deep learning techniques such as an enhanced version of super-resolution reconstruction (SR) are used to enhance the details and clarity of the image while improving the resolution. Useful feature information such as edges, textures, shapes, etc. are extracted from the modified data, and deep learning models such as convolutional neural networks (CNNs) are used to automatically learn and extract high-level feature representations.
[0045] In an embodiment of the present invention, the machine learning model includes but is not limited to decision trees, random forests, support vector machines, neural networks; the deep learning network includes but is not limited to convolutional neural networks (CNNs), generative adversarial networks (GANs), recurrent neural networks (RNNs) and their variants.
[0046] In an embodiment of the present invention, the processing configuration in the resolution modification step is dynamically adjustable, and the processing parameters are adjusted or the processing algorithm is switched in real time according to the needs of the actual application scenario or the user's instructions. Realizing the dynamic adjustability of the processing configuration is the key to ensuring the flexibility and adaptability of data processing. This dynamic adjustment capability allows the system to optimize the processing parameters or switch to the most appropriate processing algorithm in real time according to the specific needs of the actual application scenario or the user's immediate instructions to achieve the best processing effect. For example, in high-definition video surveillance, a higher resolution may be required to capture more details; while in a remote driving assistance system, a lower resolution may be selected to reduce the amount of data considering the limitation of the data transmission bandwidth. When adjusting the resolution, the parameters of the scaling algorithm (such as bilinear interpolation) used can also be dynamically adjusted, including the size, shape and weight of the interpolation kernel. In some cases, the system can also automatically adjust the processing configuration based on historical data and statistical information to adapt to changing application scenarios and needs.
[0047] In an embodiment of the present invention, the data acquisition device is a camera, the acquired raw data is image data, and the resolution of the raw image data is configured as a first resolution, such as a first resolution size of 1920*1080;
[0048] Downsampling the original image data to downsample the image resolution to a second resolution, where the second resolution is lower than the first resolution, such as the second resolution is 640*360;
[0049] Use bilinear interpolation to upsample and restore the image resolution to the first resolution;
[0050] The restored image data is added to the data sample to increase the data volume. When processing data samples, the sample volume can be directly expanded, or when training the algorithm, the algorithm can be integrated into the data enhancement algorithm to improve the generalization ability of the trained algorithm on different hardware.
[0051] In an embodiment of the present invention, downsampling is performed by using the average of the nearest pixels of a block with a length and width of three pixels. Specifically, the original image captured by the camera is divided into a plurality of blocks with a length and width of three pixels, and for each divided block, the average of all pixel values inside it is calculated. Since the block size is 3x3, each block contains 9 pixels. The grayscale values of these pixels (or the values of each color channel for a color image) are added and then divided by 9 to obtain the average of the block. The calculated average of each block is used as the new pixel value of the corresponding position in the downsampled image. Since each original block is simplified to one pixel, the downsampled image will be 1 / 3 of the original image in width and height. This downsampling method helps to reduce high-frequency details in the image while retaining a certain image structure and content.
[0052] In the embodiment of the present invention, the data acquisition device is a radar, the acquired raw data is raw radar frame data, the raw radar frame data is configured with a first line number, and the first line number is 80 line bundles;
[0053] By deleting data of a part of fixed line bundles in the original radar frame data, the original radar frame data is modified to a second line number, wherein the second line number is lower than the first line number, and the second line number is 64 line bundles;
[0054] The modified radar frame data is added to the data sample to increase the data volume. The sample volume can be directly expanded during data sample processing, or the algorithm can be integrated into the data enhancement algorithm during algorithm training to improve the generalization ability of the trained algorithm on different hardware.
[0055] The present invention also provides an embodiment of a system for increasing data quantity by modifying data quality, comprising:
[0056] Data acquisition module: acquires raw data from one or more data sources through a preset data acquisition device; the raw data includes but is not limited to images, videos, audio, text, and sensor signals;
[0057] Resolution modification module: using one or more combinations of machine learning models, deep learning networks, and reinforcement learning algorithms to perform resolution modification on the original data obtained in the data acquisition step;
[0058] Result acquisition module: After the resolution modification process is completed, the data after the resolution modification is obtained and saved.
[0059] It should be noted that the information interaction, execution process and other contents between the above-mentioned modules are based on the same concept as the method embodiment of the present application, and are a system corresponding to the above-mentioned method of increasing data volume by modifying data quality. All implementation methods in the above-mentioned method embodiment are applicable to the embodiment of the device. Its specific functions and technical effects brought about can be found in the method embodiment part, which will not be repeated here.
[0060] The present invention also provides an embodiment of an electronic device, the electronic device comprising: a processor and a memory;
[0061] The memory stores instructions executable by the processor;
[0062] When the processor is configured to execute the instructions, the electronic device implements the above-mentioned method of increasing data volume by modifying data quality.
[0063] The present invention also provides an embodiment of a computer-readable storage medium, the computer-readable storage medium comprising: computer software instructions;
[0064] When the computer software instructions are executed in an electronic device, the electronic device implements the above-mentioned method of increasing data quantity by modifying data quality.
[0065] In addition, the present invention can use a trained algorithm model with strong generalization ability to pre-label new sample data, thereby achieving rapid growth of sample data, and can support sample data of multiple resolutions; it can continuously improve the adaptability of the algorithm model to scenarios, types of targets, and image data of multiple manufacturers and resolutions.
[0066] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.
Claims
1. A method for increasing the amount of data by modifying the quality of the data, It is characterized in that The following steps are involved: Data acquisition step: acquiring raw data from one or more data sources through a preset data acquisition device; the raw data includes but is not limited to images, videos, audio, text, and sensor signals; Resolution modification step: using one or more combinations of machine learning models, deep learning networks, and reinforcement learning algorithms to perform resolution modification processing on the original data obtained in the data acquisition step; Result acquisition step: After the resolution modification process is completed, the data after the resolution modification is obtained and saved.
2. The method for increasing data quantity by modifying data quality according to claim 1, It is characterized in that Also includes: Optimization and verification steps: Perform quality assessment and optimization processing on the acquired data after the modified resolution, including but not limited to denoising, enhancement and feature extraction. At the same time, through the preset verification mechanism, evaluate the effect of the data after the modified resolution in practical applications.
3. The method for increasing data quantity by modifying data quality according to claim 1, It is characterized in that The machine learning models include but are not limited to decision trees, random forests, support vector machines, and neural networks; the deep learning networks include but are not limited to convolutional neural networks (CNNs), generative adversarial networks (GANs), recurrent neural networks (RNNs) and their variants.
4. The method for increasing data quantity by modifying data quality according to claim 1, It is characterized in that The processing configuration in the resolution modification step is dynamically adjustable, and the processing parameters are adjusted or the processing algorithm is switched in real time according to the requirements of the actual application scenario or user instructions.
5. The method for increasing data quantity by modifying data quality according to claim 1, It is characterized in that The data acquisition device is a camera, the acquired raw data is image data, and the resolution of the raw image data is configured as a first resolution; Downsampling the original image data to downsample the image resolution to a second resolution, where the second resolution is lower than the first resolution; Use bilinear interpolation to upsample and restore the image resolution to the first resolution; Add the restored image data to the data sample.
6. The method for increasing data quantity by modifying data quality according to claim 5, It is characterized in that Downsampling is performed by taking the average of the nearest pixels in a block with a length and width of three pixels.
7. The method for increasing data quantity by modifying data quality according to claim 1, It is characterized in that The data acquisition device is a radar, the acquired raw data is raw radar frame data, and the raw radar frame data is configured with a first line number; By deleting data of a part of fixed line bundles in the original radar frame data, the original radar frame data is modified to a second line number, where the second line number is lower than the first line number; Add the modified radar frame data to the data sample.
8. A system for increasing the quantity of data by modifying its quality, It is characterized in that include: Data acquisition module: acquires raw data from one or more data sources through a preset data acquisition device; the raw data includes but is not limited to images, videos, audio, text, and sensor signals; Resolution modification module: using one or more combinations of machine learning models, deep learning networks, and reinforcement learning algorithms to perform resolution modification on the original data obtained in the data acquisition step; Result acquisition module: After the resolution modification process is completed, the data after the resolution modification is obtained and saved.
9. An electronic device, It is characterized in that The electronic device comprises: a processor and a memory; The memory stores instructions executable by the processor; When the processor is configured to execute the instructions, the electronic device implements the method for increasing data volume by modifying data quality as described in any one of claims 1 to 7.
10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium includes: computer software instructions; When the computer software instructions are executed in an electronic device, the electronic device implements the method for increasing data quantity by modifying data quality as described in any one of claims 1 to 7.