Neural Network Plausibility Tester for False Detection Prevention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks used in camera-based driver assistance systems for vehicles are prone to false object detections and classification errors, which can lead to safety issues due to their vulnerability to deception and high confidence predictions for unrecognizable images.
Innovation Solution
A method and system that involves reading results and confidence values from neural network layers, using a plausibility tester to check the correctness of object detection by comparing confidence values with predefined threshold values, and preventing false outputs to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for object detection, then detection capability is improved, but vulnerability to false detections increases
Solution Approach 1:
A plausibility tester is introduced as an intermediary component between the neural network and the output stage. This mediator checks the confidence values generated by the neural network against predefined plausibility criteria, blocking false detections while allowing valid detections to pass through. The plausibility tester acts as a filter that resolves the contradiction by adding a verification layer without modifying the core detection network.
Solution Approach 2:
The system implements feedback by using the confidence values from the neural network as input to the plausibility tester, which then determines whether to accept or reject the detection result. This feedback mechanism allows the system to self-regulate by comparing the neural network's confidence against external plausibility standards, thereby reducing false detections while maintaining detection capability.
2Reliability
If confidence values are used for verification, then false positives are reduced, but system complexity increases
Solution Approach 1:
The plausibility tester uses simple, computationally inexpensive confidence value comparisons rather than complex verification models. The confidence values themselves are already generated by the neural network, so the verification process reuses existing data with minimal additional computational overhead. This approach reduces false positives without significantly increasing system complexity.
Solution Approach 2:
The confidence values generated by the neural network serve multiple functions: they represent the detection output and simultaneously serve as the verification criterion for the plausibility tester. This multi-functionality eliminates the need for separate verification data or additional complex validation mechanisms, reducing system complexity while maintaining reliability.
3Measurement precision
If all layer results are checked, then detection accuracy is improved, but computational overhead increases
Solution Approach 1:
Instead of checking all layer results, the system extracts only the confidence values from the neural network output for verification. This extraction approach focuses computational resources on the most critical verification parameter (confidence) rather than processing all intermediate layer results, thereby maintaining detection accuracy while reducing computational overhead and energy consumption.
Data Source
AI summary
The present invention relates to a computer-implemented method and a system for testing the output of a neural network (1) having a plurality of layers (11), which detects or classifies objects. The method comprises the step (S1) of reading at least one result from at least one first layer (11) and the confidence value thereof, which is generated in the first layer (11) of a neural network (1), and the step (S2) of checking a plausibility of the result by taking into consideration the confidence value thereof so as to conclude whether the object detection by the neural network (1) is correct or false. The step (S2) of checking comprises comparing the confidence value for the result with a predefined threshold value. In the event that it is concluded in the checking step (S2) that the object detection is false, output of the object falsely detected by the neural network is prevented.