Search

KR-20260063584-A - Method And Apparatus for Generating Training Data for Lightweighting Artificial Intelligence Model

KR20260063584AKR 20260063584 AKR20260063584 AKR 20260063584AKR-20260063584-A

Abstract

A method and apparatus for generating training data for lightweighting an artificial intelligence model are disclosed. According to one aspect of the present disclosure, a computer implementation method for generating training data for lightweighting an artificial intelligence model is provided, comprising: a process of acquiring time-series data measured using an optical transceiver; a process of setting a monitoring section among the time-series data based on a maximum distance at which the optical transceiver can detect an object; a process of setting an analysis target section among the monitoring section based on the amount of light incident on the optical transceiver; and a process of generating the training data based on the time-series data included in the analysis target section.

Inventors

  • 곽재영

Assignees

  • 에스케이텔레콤 주식회사

Dates

Publication Date
20260507
Application Date
20241030

Claims (10)

  1. As a computer implementation method for generating training data for lightweighting an artificial intelligence model, The process of acquiring time series data measured using an optical transceiver; A process of setting a monitoring interval among the time series data based on the maximum distance at which the optical transceiver can detect an object; A process of setting an analysis target section among the monitoring sections based on the amount of light incident on the optical transceiver; and A computer implementation method comprising a process of generating training data based on time series data included in the above-mentioned analysis target interval.
  2. In paragraph 1, The process of setting the monitoring interval among the above time series data is, The process includes selecting a portion of the time corresponding to the time from the first time point to the second time point among the entire intervals included in the above time series data, The above first point in time is the point in time when light is emitted from the optical transceiver, and The above second point in time is a computer implementation method in which, after light is emitted from the light transceiver, it is reflected by the object furthest from the light transceiver among the objects detectable by the light transceiver and then incident back on the light transceiver.
  3. In paragraph 2, A computer implementation method wherein the second time point is calculated by adding the effective detection time, which is calculated based on the maximum distance at which the optical transceiver can detect an object, to the first time point.
  4. In paragraph 2, The process of setting the analysis target section among the above monitoring sections is, The process includes selecting a portion of the entire section included in the above monitoring section that corresponds to the time from the third point in time to the fourth point in time, The third and fourth points in time are the points in time when, after light is emitted from the optical transceiver, it is reflected by an object located at a certain distance from the optical transceiver and then incident back on the optical transceiver. A computer-implemented method in which the above constant distance is calculated based on the number of photons contained in the light incident on the light transceiver at the third time point and the fourth time point.
  5. In paragraph 4, The above third time point is calculated by adding the first round-trip time calculated based on the distance between the optical transceiver and the object estimated when the number of photons is 1,000 at the above first time point, and A computer implementation method wherein the fourth time point is calculated by adding a second round-trip time calculated based on the distance between the optical transceiver and the object estimated when the number of photons is 10 at the first time point.
  6. In paragraph 1, The above time series data is data that records the waveform of light when light emitted from the optical transceiver at a constant time period is reflected by an object and incident back on the optical transceiver. The above-mentioned training data is a computer implementation method generated based on a plurality of analysis target intervals generated based on time series data collected over a plurality of the above-mentioned periods.
  7. In paragraph 1, The above artificial intelligence model is, Based on the input data, predict the probability for each of the predefined weather conditions, and A computer implementation method for classifying weather conditions near the optical transceiver into one of predefined weather conditions based on the above probability.
  8. In Paragraph 7, The above training data includes time series data and Ground Truth (GT) weather conditions included in the above analysis target interval, and The above GT weather conditions are a computer implementation method in which time series data included in the above analysis target interval is labeled.
  9. As a device for generating training data for lightweighting artificial intelligence models, Memory for storing instructions; and at least one processor, comprising The above at least one processor executes the above instructions, Time series data measured using an optical transceiver is acquired, and Based on the maximum distance at which the optical transceiver can detect an object, a monitoring interval is set among the time series data, and Based on the amount of light incident on the above optical transceiver, the section to be analyzed among the above monitoring sections is set, and A device that generates the training data based on time series data included in the above-mentioned analysis target interval.
  10. A computer program stored on a computer-readable recording medium to execute each process included in the method according to any one of paragraphs 1 through 8.

Description

Method and Apparatus for Generating Training Data for Lightweighting Artificial Intelligence Model The present disclosure relates to a method and apparatus for generating training data for lightweighting an artificial intelligence model. The following description merely provides background information related to the present embodiment and does not constitute prior art. A technology for detecting the distance between an optical transceiver and an object using an optical transceiver transmits an optical pulse using a laser mounted on the optical transceiver, and when the optical pulse is reflected by a target and the light is received by the optical transceiver, the waveform of the received light is processed to calculate the distance between the optical transceiver and the target. In this regard, noise may occur in the light received by the optical transceiver depending on the weather conditions between the optical transceiver and the target. For example, noise can be generated due to light scattering caused by fog particles. Consequently, it is necessary to consider noise caused by weather conditions when converting the waveform of the received light into a digital signal. To address this problem, methods and devices have been devised to process noise by analyzing weather conditions based on the waveform of the received light using Artificial Intelligence (AI) models, including machine learning. AI models can learn the noise that may occur under various weather conditions and perform accurate signal processing based on this. Recently, technologies are being developed to use these AI models in edge devices to analyze weather conditions and process noise in real time. Edge computing technology processes data on local devices, unlike the cloud, and has the advantage of reducing network bandwidth and improving real-time responsiveness. However, because edge devices generally have limited computing resources (CPU, memory, storage), it is difficult to efficiently run complex artificial intelligence models. In particular, computationally intensive tasks, such as processing time-series data through optical transceivers, can lead to performance limitations of edge devices. Therefore, in order to efficiently run artificial intelligence models on edge devices, methods and devices for lightweighting artificial intelligence models are required. FIG. 1 is a block diagram schematically illustrating an exemplary system to which the present disclosure may be applied. FIG. 2 is a block diagram schematically showing an optical transceiver including an artificial intelligence model according to one embodiment of the present disclosure. FIG. 3 is a drawing for explaining an artificial intelligence model according to one embodiment of the present disclosure. FIG. 4a is a diagram illustrating a trigger signal and time series data according to one embodiment of the present disclosure. FIG. 4b is a drawing for explaining the process of setting a monitoring section according to one embodiment of the present disclosure. FIG. 4c is a drawing for explaining the process of setting the analysis target section according to one embodiment of the present disclosure. FIG. 5 is an example of a graph showing the relationship between the distance from an optical transceiver to an object and the number of photons received by the optical transceiver according to one embodiment of the present disclosure. FIG. 6 is a diagram illustrating the process of training an artificial intelligence model according to one embodiment of the present disclosure. FIG. 7 is a flowchart schematically illustrating a method for generating training data for lightweighting an artificial intelligence model according to one embodiment of the present disclosure. FIG. 8 is a schematic block diagram of an exemplary computing device that can be used to implement the devices and methods described in the present disclosure. Some embodiments of the present disclosure are described in detail below with reference to exemplary drawings. It should be noted that in assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings. Furthermore, in describing the present disclosure, if it is determined that a detailed description of related known components or functions could obscure the essence of the present disclosure, such detailed description is omitted. In describing the components of the embodiments according to the present disclosure, symbols such as first, second, i), ii), a), b), etc., may be used. These symbols are intended only to distinguish the components from other components, and the essence, order, or sequence of the components is not limited by the symbols. When a part in the specification is described as 'comprising' or 'having' a component, this means that, unless explicitly stated otherwise, it does not exclude other components but may include addi