Image Recognition Fine-Tuning for On-Site Environment Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition models struggle with limited sample types due to varied environmental factors, leading to poor learning effects and high computational demands.
Innovation Solution
A data processing method involving on-site image capture, local model fine-tuning, and synchronization with a server-based model to adapt to specific environments, enhancing recognition precision and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large quantity of images are captured in different environments to perform model training, then model recognition capability is enhanced, but computing power consumption and time consumption increase significantly
Solution Approach 1:
The patent divides model training into two segments: (1) initial model training on the server using a large quantity of images from different environments, and (2) local fine-tuning on edge devices using a small quantity of images from specific on-site environments. This segmentation allows the server to perform comprehensive training once, while edge devices perform efficient local adaptation, reducing repeated computing power consumption.
Solution Approach 2:
The patent performs preliminary model training on the server before deploying to edge devices. The pre-trained model contains general recognition capabilities learned from diverse environmental images. This preliminary action reduces the need for edge devices to process large quantities of images locally, thereby reducing computing power consumption at the edge.
2Reliability
If a large quantity of images are captured in different environments to perform model training, then model recognition capability is enhanced, but time consumption increases significantly
Solution Approach 1:
The patent segments the time-consuming training process into server-side initial training and edge-side fine-tuning. The server performs comprehensive training offline, while edge devices only perform quick fine-tuning with少量 images, dramatically reducing the time required at the edge while maintaining recognition capability.
Solution Approach 2:
The patent performs preliminary model training on the server before deployment. This pre-trained model already contains general recognition capabilities, so edge devices only need to perform quick local fine-tuning rather than training from scratch, significantly reducing time consumption at the edge.
3Adaptability or versatility
If images used for model training cover various environments, then sample types that can be learned by models are improved, but the complexity of data collection and processing increases
Solution Approach 1:
The patent segments data collection into two phases: (1) server collects diverse environmental images for initial training, and (2) edge devices only collect少量 on-site images for fine-tuning. This segmentation reduces the complexity of continuous data collection at edge devices while maintaining adaptability through the pre-trained model's exposure to various environments.
Solution Approach 2:
The patent performs preliminary data collection and model training on the server with diverse environmental images. This preliminary action ensures the model has learned various sample types before deployment, so edge devices don't need to collect and process diverse data themselves, reducing their complexity.
Data Source
AI summary
This application discloses a data processing method performed by a computer device. The method includes: transmitting K images photographed of an object to a server, where the server obtains K first prediction results by using an image recognition model; constructing a fine-tuning training set according to the K images and the K first prediction results; obtaining a second prediction result of each image in the fine-tuning training set by using a to-be-trained model; updating a model parameter of the to-be-trained model according to the second prediction result of each image and the first prediction result of the image in the fine-tuning training set, to obtain a local recognition model and a model adjustment parameter; and transmitting the model adjustment parameter to the server if a model fine-tuning condition is satisfied, so that the server updates a model parameter of the image recognition model according to a model adjustment parameter set.