Neural Network Feedback for Detecting Unknown Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object recognition models fail to detect and take action on unknown or unrecognizable objects within the input information, leading to incomplete recognition and missed detection.
Innovation Solution
A neural network mechanism that segments objects in the input information, determines the quantity of recognizable objects, compares this with the total object count, and trains the model when a discrepancy is found, allowing it to recognize and learn from unrecognizable objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a conventional object recognition model is used to detect objects in under-test information, then the model can recognize familiar objects with known training data, but it fails to detect or take action on unknown or unrecognizable objects
Solution Approach 1:
The system implements a feedback mechanism where the object recognition model's detection results are compared with segmentation results. When the quantity of recognized objects differs from the total quantity of segmented objects, the system generates feedback to train the model with unrecognizable objects, enabling continuous improvement and adaptation to unknown object types
Solution Approach 2:
The system segments objects in under-test information using a segmentation model to obtain segmentation results. This segmentation step divides the recognition task into two parts: recognizing familiar objects through the object recognition model and identifying unrecognizable objects through comparison with segmentation results, allowing the system to handle both known and unknown objects
2Measurement precision
If the object recognition model is trained with a large amount of manually labeled data to improve recognition accuracy, then the model can better recognize familiar objects, but it cannot understand or take action on objects that are unknown to it
Solution Approach 1:
The system introduces a segmentation model as an intermediary to bridge the gap between recognized and unrecognized objects. The segmentation model provides comprehensive object boundaries and quantities, allowing the system to identify unrecognizable objects without losing information about them, even when the object recognition model cannot classify them
Solution Approach 2:
The system performs preliminary segmentation of all objects in the under-test information before attempting recognition. This preliminary action ensures that no object information is lost, and the segmentation results serve as a reference for identifying which objects remain unrecognizable after the recognition process
Data Source
AI summary
A comprehension method of neural network for unrecognizable objects comprises: determining at least one type of at least one object in an under-test information by an object recognition model to output a quantity value of recognizable objects, generating a total quantity value of objects by segmenting the at least one object in the under-test information, and comparing whether the quantity value of the recognizable objects is equal to the total quantity value of the objects. When the quantity value of recognizable objects is less than the total quantity value of objects, it means the under-test information contains unknown objects that are not recognized by the object recognition model. The processing unit trains the object recognition model that at least one unrecognizable object is included in the under-test information. The present invention provides a neural network mechanism for comprehending unknown objects similar to humans.