Object Recognition Model Training Using Metadata Embedding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition models have relatively low accuracy in recognizing various objects such as plants and insects, necessitating an improved training method that incorporates metadata for enhanced recognition accuracy.
Innovation Solution
The proposed object recognition model training method involves acquiring a training set with input samples including object images and metadata, normalizing the metadata, and using a neural network to train the model by generating image feature vectors, embedding feature vectors from metadata, fusing these vectors, and achieving classification results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object recognition models are trained using only image features, then the training process is simple, but the recognition accuracy is low
Solution Approach 1:
The patent combines image features and metadata features into a unified feature representation. The metadata embedding module converts metadata into feature vectors that are fused with image features through concatenation or addition, creating a composite feature vector that leverages both visual and contextual information for improved recognition accuracy
Solution Approach 2:
The patent introduces an embedding module as an intermediary component that transforms metadata into a format compatible with image features. This embedding module acts as a mediator that bridges the gap between unstructured metadata and structured image features, enabling their effective integration without direct complex interaction
2Measurement precision
If metadata is integrated into the training process, then recognition accuracy improves, but the training complexity increases
Solution Approach 1:
The patent divides the training process into distinct modules: an image processing branch that extracts features from images, and a metadata embedding branch that processes metadata separately. These segmented processing paths can be trained independently and then combined, simplifying the overall training process while maintaining the benefits of integrated features
Solution Approach 2:
The patent performs preliminary processing of both images and metadata before their fusion. Image preprocessing includes normalization and augmentation, while metadata undergoes embedding transformation. This preliminary action prepares both data types in advance, making the subsequent fusion and training processes more efficient and manageable
3Measurement precision
If only image features are used for object recognition, then the processing is fast, but the recognition precision is insufficient
Solution Approach 1:
The patent extracts only the most relevant features from metadata through the embedding module, rather than processing all raw metadata information. This selective extraction of essential metadata features reduces the dimensional complexity while preserving the information most useful for recognition, thereby maintaining processing efficiency
Solution Approach 2:
The patent transforms metadata from its original format into a different dimensional representation through embedding. This dimensionality transformation converts categorical or text-based metadata into continuous feature vectors that match the dimensional space of image features, enabling efficient fusion without significant computational overhead
Data Source
AI summary
The disclosure relates to a training method of an object recognition model, an object recognition method, and an object recognition device. The training method includes: acquiring a training set, in which the training set includes a plurality of groups of input samples for training and marking results corresponding to the input samples, each input sample includes an object image and metadata, and the metadata is configured to describe a corresponding object image; on the basis of a neural network, training an object recognition model by using the training set; and ending the training when a training accuracy is greater than or equal to a preset accuracy, and obtaining a trained object recognition model.