Minimal-Entropy Sensor Allocation for Faster Object Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor allocation strategies for object classification are non-optimal, leading to inefficient resource utilization and longer times to reach desired classification thresholds, degrading overall performance due to a lack of consideration for the information gained from each observation.
Innovation Solution
An optimal observation strategy is determined using Bayesian probability and information theory to minimize expected categorical posterior entropy, guiding sensor allocation to maximize information gain and reduce the number of dwells required for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If non-optimal observation strategy is used for sensor allocation, then resource management is simpler, but classification time increases and performance degrades
Solution Approach 1:
The patent applies parameter changes by using Bayesian probability updates to dynamically adjust the state representation of targets based on sensor observations. The entropy calculation uses probability parameters to guide sensor allocation, transforming the resource management problem into a mathematical optimization based on information theory parameters.
Solution Approach 2:
The patent implements feedback through the Bayesian update mechanism where sensor observations are continuously fed back to update target state probabilities. The entropy calculation provides feedback on the current level of uncertainty, which then guides the next sensor allocation decision, creating a closed-loop system that adapts to new information.
2Loss of time
If non-optimal sensor allocation is used, then fewer computational resources are required, but more dwells are needed to reach classification thresholds
Solution Approach 1:
The patent applies preliminary action by pre-calculating entropy values for potential sensor allocations based on current probability states. This allows the system to identify optimal sensor assignments in advance before actual observations are made, reducing the number of dwells needed while the computational overhead is spread across pre-computation rather than real-time decision-making.
3Measurement precision
If optimal observation strategy based on minimal entropy is used, then classification efficiency improves, but computational complexity increases
Solution Approach 1:
The patent replaces mechanical or heuristic sensor allocation methods with an information-theoretic approach based on entropy minimization. Instead of using rule-based or trial-and-error allocation, the system uses mathematical entropy calculations to determine optimal sensor assignments, substituting computational mathematics for mechanical decision-making processes.
Data Source
AI summary
Systems, devices, methods, and computer-readable media for sensor allocation and object classification. A method can include determining, by a sensor allocator, a first dwell assignment, the first dwell assignment associating a sensor of sensors with a respective object of objects to be classified, providing, by the sensor allocator, the first dwell assignment to one or more devices capable of orienting the sensors to measure corresponding assigned objects, receiving, from a classifier, that classified the objects based on measurements from the sensors in implementing the first dwell assignment, a classification for each of the objects, determining, by the sensor allocator and based on the classification, a second dwell assignment associating each sensor of the sensors with a different object of the objects than that assigned in the first dwell assignment, and providing, by the sensor allocator, the second dwell assignment to the one or more devices.