Deep Learning Web Traffic Recognition via Hexadecimal Image Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning methods struggle to recognize diverse character sets in web traffic (URLs) due to limitations in character recognition, such as only recognizing 68 characters, and are unable to continuously learn new data, leading to decreased accuracy over time and difficulty in handling changes in website structures.
Innovation Solution
The method employs hexadecimal image conversion and incremental learning, where multiple deep learning machines learn and overlap data using UTF-8 hexadecimal encoding, allowing recognition of a broader character set and continuous learning, with weights adjusted based on error rates to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the deep learning module is configured to recognize only 68 characters (26 English small letters, 10 numbers, and 32 symbols), then the device complexity is reduced and ease of operation is improved, but the adaptability to recognize diverse character sets in web traffic (URLs) deteriorates
Solution Approach 1:
The patent changes the character encoding parameter from a limited 68-character set to UTF-8 hexadecimal encoding, which can represent any character in the world's languages. This parameter change enables the deep learning module to recognize diverse character sets in web traffic while maintaining operational simplicity through automated encoding conversion.
2Ease of manufacture
If the deep learning module uses a fixed character set of 68 characters, then the manufacturing precision of the recognition system is simplified, but the measurement precision of web traffic recognition deteriorates
Solution Approach 1:
The patent changes the encoding parameter from a fixed 68-character set to UTF-8 hexadecimal encoding, expanding the character recognition capability to include all languages and character types found in web traffic, thereby improving measurement precision without significantly complicating the manufacturing process.
Solution Approach 2:
The patent introduces hexadecimal encoding as an intermediary layer between the input web traffic and the deep learning module. This intermediary converts diverse character sets into a standardized hexadecimal format that the deep learning module can process, thereby improving recognition precision while maintaining ease of manufacture.
3Stability of the object's composition
If the deep learning module is trained on predetermined data and cannot learn new data, then the stability of the learning system is improved, but the adaptability to new web traffic patterns deteriorates
Solution Approach 1:
The patent implements incremental learning that allows the deep learning module to dynamically update its knowledge base by continuously learning new web traffic patterns while retaining previously learned patterns. This dynamic learning capability enables the system to adapt to new web traffic types without losing stability in previously learned recognition patterns.
4Loss of time
If the deep learning module learns only static data, then the loss of time for learning is reduced, but the productivity of recognizing new web traffic patterns deteriorates
Solution Approach 1:
The patent implements continuous incremental learning where the deep learning module continuously learns new web traffic patterns in an ongoing manner rather than through periodic retraining. This continuous learning process maintains high productivity in recognizing new patterns while minimizing time loss by integrating learning into the operational flow.
Data Source
AI summary
Provided is a deep learning method including a step of each of at least two or more deep learning machines learning a web traffic by using a hexadecimal; a step of the deep learning machines learning the web traffic by using an incremental learning using a weight; a step of, when the web traffic is received, each of the deep learning machines encoding a character string of the web traffic with UTF-8 hexadecimal; a step of each of the deep learning machines converting the character string into an image and deep learning the image.