Recognition Device Using Feature Maps for Residue Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semantic segmentation technologies, such as U-Net, face challenges in achieving high recognition accuracy when recognizing difficult areas in images, particularly in cases where the target area includes residues not corresponding to leftovers, like liquid or food remnants, which can be mistaken for leftovers.
Innovation Solution
A recognition device and model learning device that incorporate related information, such as menu item details, to enhance image recognition by converting images into feature maps and calculating these maps using latent information from the related data, allowing for accurate division of areas like containers, targets, and non-targets, and updating model parameters to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only image information is used for semantic segmentation, then the device complexity is low, but the recognition accuracy for difficult areas deteriorates
Solution Approach 1:
The patent combines image information with related information (menu items, preparation methods, ingredients) to create comprehensive recognition target data. This merging of multiple information sources enables the model to achieve high recognition accuracy for difficult areas by utilizing both visual features and contextual knowledge from the related information.
Solution Approach 2:
The patent introduces a model as an intermediary that processes and integrates image information with related information. The model learns to combine these different types of data through training, acting as a mediator that transforms multiple input sources into accurate segmentation results without requiring direct complex processing in the recognition stage.
2Measurement precision
If semantic segmentation is applied using only image information, then the processing speed is fast, but the recognition accuracy for difficult areas deteriorates
Solution Approach 1:
The patent performs model training in advance where the model learns to effectively combine image information with related information. This preliminary action during the training phase enables fast and accurate recognition during actual use, as the integration logic has already been learned and stored in the model parameters.
3Measurement precision
If the model processes both image information and related information, then the recognition accuracy improves, but the data processing complexity increases
Solution Approach 1:
The patent designs a universal model structure that can handle both image information and related information through a unified processing framework. The model accepts multiple types of input data and processes them together to produce segmentation results, eliminating the need for separate processing pipelines for different data types.
Data Source
AI summary
A recognition device includes a data extraction unit, a recognition unit, and a ratio estimation unit. The data extraction unit acquires related information related to a target in a recognition target image which is a post-image obtained through photographing before and after treatment on a container that stores the target, and extracts recognition target data which is a combination of the recognition target image and the related information The recognition unit accepts the recognition target data as an input to a model learned in advance and outputs a recognition result obtained by recognizing an area where at least the container, the target, and a portion other than the target are divided by an output of the model. The ratio estimation unit estimates a ratio of the target in the recognition target image based on the recognition result and an area ratio in a pre-stored pre-image obtained through the photographing before and after the treatment. The model recognizes the area by converting the recognition target image into a feature amount map and calculating the feature amount map in a weighting manner by latent information obtained from the related information.