迁移方法、迁移装置、计算机设备和计算机可读存储介质
By identifying and weakening the weights of low-quality sensors, the model accuracy problem caused by the inconsistency in the types and quantities of sensors is solved, and efficient transfer and high-precision recognition of deep learning models in the target environment are achieved.
CN116304671BActive Publication Date: 2026-07-17GD MIDEA AIR CONDITIONING EQUIP CO LTD +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GD MIDEA AIR CONDITIONING EQUIP CO LTD
- Filing Date
- 2022-12-28
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In home environments, the types and quantities of sensors vary across different user environments, leading to low-quality sensors affecting the accuracy of collaborative sensing models and making it difficult to effectively transfer deep learning models.
Method used
Low-quality sensors in the target environment are identified and assigned weakened weights. The weights of each sensor are fixed, and the target deep learning model after transfer is determined based on the dataset to be transferred and the deep learning model in the source environment.
Benefits of technology
Reduce the impact of low-quality sensor data on the target deep learning model, accelerate training convergence speed, and ensure high recognition accuracy in the target environment.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116304671B_ABST
Abstract
本发明公开了一种迁移方法、迁移装置、计算机设备和计算机可读存储介质。迁移方法包括：识别出目标环境下当前传感器中的低质量传感器并为低质量传感器分配弱化权重，弱化权重小于预设权重；固定每个当前传感器的权重，基于待迁移数据集和源环境下的深度学习模型确定迁移后的目标深度学习模型，以将深度学习模型从源环境迁移到目标环境。本发明的技术方案，通过识别当前传感器中的低质量传感器并为低质量传感器分配较小权重，可以极大降低低质量传感器数据对目标深度学习模型的影响，加快目标环境下目标深度学习模型的训练收敛速度，确保目标深度学习模型能够达到较高的识别精度。
Need to check novelty before this filing date? Find Prior Art