Commodity Recognition Apparatus Foreign Object Exclusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object recognition technologies for commodities like vegetables and fruits face challenges in maintaining high recognition rates due to variations in texture and color based on producing areas, and are inefficient in excluding foreign objects from the learning process, requiring manual confirmation by operators.
Innovation Solution
A commodity registration apparatus with a processor that extracts feature values from captured images, compares them to stored values, generates relationship information, and excludes images meeting certain conditions to improve the learning process by automatically excluding foreign objects and updating feature values for accurate recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the learning function is provided to newly register or update the feature value from images that are newly captured, then the recognition rate is improved, but if a foreign object such as a hand of an operator is contained in a captured image for learning, the recognition rate is lowered
Solution Approach 1:
The system performs preliminary actions by extracting feature values from captured images before the learning process, and uses these pre-extracted features to determine whether to include the image in learning. The processor extracts feature values, calculates similarity degrees with reference images, and makes determination about image inclusion based on these pre-calculated metrics, preventing foreign objects from contaminating the learning process.
Solution Approach 2:
The system introduces an intermediary mechanism using similarity degree calculation as a filter. The processor calculates the similarity degree between extracted feature values and reference image feature values, and uses this intermediate metric to determine whether to include the captured image in learning. This intermediary similarity assessment acts as a gatekeeper to prevent foreign objects from entering the learning process while allowing valid commodity images to be processed.
2Reliability
If the operator confirms states of the captured images one by one to determine whether or not the captured image is set as the target of the learning operation, then the foreign object exclusion is improved, but the operation becomes very troublesome and inefficient
Solution Approach 1:
The system performs self-service by automatically determining whether captured images should be used for learning based on the extracted feature values and calculated similarity degrees. The processor autonomously evaluates each captured image, compares it with reference images, and makes the determination about inclusion in learning without requiring operator intervention. This eliminates the troublesome manual confirmation process while maintaining reliable foreign object exclusion.
Solution Approach 2:
The system replaces the mechanical manual confirmation process with an automated computational system. Instead of operators visually inspecting and manually selecting images, the processor automatically extracts features, calculates similarities, and makes inclusion determinations through computational algorithms. This substitution of mechanical manual operation with automated computational processing significantly reduces operator workload while maintaining or improving exclusion accuracy.
3Measurement precision
If the feature value is extracted from captured images and compared with stored feature values, then the recognition accuracy is improved, but the processing time increases
Solution Approach 1:
The system applies partial action by extracting and processing only the necessary feature values from captured images rather than analyzing the entire image data. The processor extracts specific feature characteristics, compares only these extracted features with stored reference values, and makes recognition determinations based on this partial analysis. This partial processing approach maintains sufficient recognition accuracy while significantly reducing processing time compared to full image analysis.
Data Source
AI summary
A commodity registration apparatus configured to perform object recognition includes an interface connected to receive captured images, a storage unit storing a dictionary for the object recognition, and a processor. The processor is configured to designate a learning target article for learning processing, extract, from each captured image, feature value indicating feature of an article contained in the captured image, compare each of the extracted feature values with stored feature values of the learning target article registered in the dictionary and calculate a similarity degree therebetween, generate relationship information indicating a relationship between the captured images based on the calculated similarity degrees, exclude captured images that meet a predetermined condition based on the relationship information, and execute the learning processing by adding, to the dictionary with respect to the learning target article, the feature values indicating features of the article contained in the non-excluded captured images.