As a cutting-edge technology, large models have powerful computing power and complex data processing capabilities, but how feasible is the model in the face of large-scale data and high-complexity tasks? This article hopes to provide you with a comprehensive perspective and reveal the potential of large AI models in theory and practice.
With the rapid development of information technology, artificial intelligence (AI) has gradually moved from science fiction to reality, and has become the core force driving the transformation of modern society.
As a cutting-edge technology in this field, AI large models are leading the trend of technological innovation with their powerful computing power and complex data processing capabilities.
However, how feasible and effective are these models in the face of large-scale data and high-complexity tasks?
This article hopes to provide you with a comprehensive perspective and reveal the potential of large AI models in theory and practice.
So what we need to do today is to carry out the expiration of certificates based on the large model, so as to conceive and select the implementation ideas.
First, the goal
Determine whether the ID card is expired based on the image of the ID card.
2. OCR technology
OCR (Optical Character Recognition) technology can convert the text of various bills, newspapers, periodicals, books, manuscripts and other printed materials into image information by scanning, and use text recognition technology to convert image information into usable text. This technology can help us extract valid text information from the ID card image for subsequent expiration judgment.
OCR: OCR technology is the abbreviation of optical character recognition (Optical Character Recognition), which is a computer input technology that converts the text of various bills, newspapers, books, manuscripts and other printed materials into image information through optical input methods such as scanning, and then uses text recognition technology to convert image information into text that can be used - scanning pictures to text.
Step 1: Prepare the dataset
Dataset Preparation - In daily life, we first collect the people around us.
#手机内容: ID card, residence permit, and other identity information that meets the requirements of the scenario: such as ID card, driver's license, passport, residence permit, and Hong Kong and Macao pass.
#收集数量: 10-30 sheets
"After we have collected enough data, the next step is to aggregate the data for subsequent processing and analysis."
Step 2: Data Aggregation
The data summary is convenient for us to evaluate and apply the data quality in the future, and make an Excel table in the following format
First of all, we can upload the image to the cloud disk and get the image link. The second column is the OCR result copied and pasted
Get in.
Step 3: OCR extraction
OCR 内容提取可行性验证，并不需要开发代码有界面，可操作交互的 OCR 产品即可腾讯 OCR：https://cloud.tencent.com/product/ocr
Step 4: The big oracle model has the ability to extract the key information required
OpenAI's platform example: https://platform.openai.com/playground?mode=chat can choose other large language models, which require that prompts can be set, and models and parameters can be adjusted.
Note: "Prompt engineering is a method of designing and improving AI prompts to improve the performance of AI. ”
Prompt Engineering is an artificial intelligence (AI) technology that improves the performance of AI by designing and improving its prompts. The goal of Prompt Engineering is to create highly effective and controllable AI systems that can perform specific tasks accurately and reliably.
Because human language is fundamentally imprecise, machines are not yet able to understand what humans are saying very well, which is why prompt engineering is a technology.
If you are a product designer or R&D person, you can use it to design and improve the prompts of the AI system, so as to improve the performance and accuracy of the AI system and bring a better AI experience to users.
Step 5: Iterative tuning
Enter the OCR content—view the result—adjust the Prompt—adjust the model/parameters—and finally output all in a unified manner.
Step 6: Complete the data
Add the output result of the third column, and manually judge whether the fourth column is True or False
Step 7: Overall effect evaluation
Aggregate sample statistics: 20 samples
1. Precision: 95%
2. Rounding rate: 100
Precision: The number of samples that are actually correct for a positive prediction
Recall: How many of the actual positive samples were predicted back
For example: there are a lot of dogs and cats in front of you, throw a bone out and hope that the dog will run over, and the accuracy is how many of the animals that run back are just dogs, and 10 run back, of which there are 8 dogs, and the accuracy rate is 80% There are 20 dogs in all the animals, and the recall rate is 8/20.
To sum up, through the "Feasibility Verification of AI-Large Model", we have deeply discussed how to use AI large model and OCR technology to determine whether the ID card is expired. The whole process includes data collection, OCR extraction, model application and result evaluation, demonstrating the excellent performance of large AI models in handling complex data tasks.
In practical applications, we have successfully achieved high-precision ID card expiration judgment. In the 20 samples tested, we achieved 95% precision and 100% recall, fully demonstrating the potential of AI large models. For example, through OCR technology, we accurately extract the expiration date information on the ID card, and through the analysis of the large language model, we can accurately determine whether the ID card is expired.
However, there were some challenges along the way. For example, some OCR results have misidentification issues, especially when dealing with blurry or low-quality images. In addition, the settings of the Prompt project and the tuning of model parameters need to be continuously tried and optimized to obtain the best judgment results. In the face of these challenges, we have gradually improved the accuracy and reliability of the system through continuous iteration and improvement.
Still, we need to continue to optimize the technology, address deficiencies in image quality and recognition accuracy, and strengthen ethical scrutiny and safety testing of AI models. In the future, with the further development of technology, AI large models are expected to achieve breakthroughs in more fields, bringing more innovation and convenience.
This article was posted by @李雪亮 on Everyone is a Product Manager and is not allowed to be reproduced without permission.
Image from Unsplash, based on the CC0 license
The views in this article only represent the author's own, everyone is a product manager, and the platform only provides information storage space services.