Cash Handling Machine Neural Network Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cash handling machines face inefficiencies in recognizing and sorting multiple denominations and currencies due to the high computational burden of large neural networks, which require frequent retraining and redeployment when currency designs change.
Innovation Solution
Implementing a series of smaller neural networks, each trained to recognize specific subsets of cash types, allowing for staged recognition and reducing computational intensity, with only the necessary networks being triggered for each article, and enabling modular updates to affected stages when currency designs change.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large neural network is trained to recognize all possible denominations and currencies, then the recognition accuracy is improved, but the computational burden and processing time increase significantly
Solution Approach 1:
The patent divides the single large neural network into multiple smaller neural networks, each specialized for recognizing specific subsets of cash types (e.g., different currencies or denominations). This segmentation allows the system to maintain high recognition accuracy for diverse cash while reducing computational burden by only activating the necessary subset of networks for each transaction.
Solution Approach 2:
The system employs partial action by selectively activating only the neural networks relevant to the current cash type being processed. Instead of running all networks simultaneously, the system activates only the necessary portion of the recognition system, thereby reducing overall computational resources consumed while maintaining comprehensive recognition capability.
2Adaptability or versatility
If a large neural network is used to recognize all cash types, then the versatility is improved, but the time to retrain and redeploy increases when currency designs change
Solution Approach 1:
The neural network system is segmented into independent modules, each responsible for specific cash types. When a new currency design is introduced, only the corresponding specific module needs to be retrained and deployed, rather than retraining the entire system. This maintains versatility while significantly reducing retraining time.
Solution Approach 2:
Each neural network module is specialized with local quality for recognizing specific cash types. This specialization means that updates to one module do not affect others, allowing targeted retraining only where needed when currency designs change, thereby preserving system versatility without the time cost of full system retraining.
3Productivity
If multiple neural networks are used to recognize different cash types, then the processing speed is improved by early termination, but the device complexity increases
Solution Approach 1:
The system segments the recognition task into multiple independent neural networks, each handling specific cash types. This segmentation enables early termination optimization where the system can stop processing once a match is found in the relevant network, significantly improving processing speed for common cash types while managing complexity through modular architecture.
Solution Approach 2:
The system dynamically selects and activates only the necessary neural networks based on the cash type being processed. This dynamic behavior allows the system to optimize processing speed by avoiding unnecessary network activations, while the modular structure keeps the overall complexity manageable through on-demand resource allocation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A cash handling machine comprising: a cash handling mechanism for receiving physical articles including cash, sensing equipment for sensing the articles, processing equipment processors, and memory storing code arranged to run on the processing equipment. The code comprises multiple neural networks arranged in a series ordered from first to last, each trained to classify which of a respective set of classes the articles belong to based on the sensed data. The set of classes of each respective neural network comprises a different respective subset of one or more respective recognized classes classifying a respective one or more types of cash recognised by the respective neural network, and further comprising an unrecognized class into which all other articles other than the respective recognized types are classified. For each article, the code starts with the first neural network in the series, using each network in turn to try to recognize the article.