Vehicle Identification Using Segmented Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning systems for identifying objects in images are computationally intensive and lack accuracy in specific tasks, such as distinguishing between different vehicle models, requiring large datasets and extensive training time.
Innovation Solution
A software application utilizing a machine learning system with a feature learning neural network trained for general tasks and a custom image classification neural network trained with a smaller dataset of specific vehicle images, employing depthwise separable convolutions and bitwise operations to reduce computational power and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional machine learning systems use traditional convolution techniques with large numbers of parameters, then they can perform general object identification tasks, but they require substantial computing power and are computationally intensive
Solution Approach 1:
The patent segments the neural network into two distinct components: a feature learning neural network trained on large general datasets, and a custom image classification neural network trained on smaller task-specific datasets. This segmentation allows the system to separate general feature extraction from specific classification tasks, reducing the computational burden on mobile devices while maintaining versatility.
Solution Approach 2:
The patent extracts and reuses the feature learning component from a pre-trained neural network, separating it from the full conventional system. This extracted feature learning network is then combined with a lightweight custom classification network, eliminating the need to run the entire conventional system on resource-constrained mobile devices.
2Adaptability or versatility
If conventional machine learning systems are trained using very large sets of images, then they can perform general identification tasks, but the training process becomes complicated and time consuming
Solution Approach 1:
The patent applies preliminary action by pre-training the feature learning neural network on large general datasets before deployment. This pre-training is performed once during system development, and the resulting feature extraction capabilities are then reused across multiple specific tasks without requiring retraining on large datasets for each new task.
Solution Approach 2:
The patent applies local quality by training only the custom image classification neural network on small, task-specific datasets rather than retraining the entire system. This allows the system to adapt to specific tasks (e.g., vehicle model identification) with minimal training data and time, while maintaining the general capabilities provided by the pre-trained feature learning network.
3Measurement precision
If conventional machine learning systems use traditional convolution techniques, then they can identify objects in images, but they require processing a large number of parameters and performing extensive mathematical operations
Solution Approach 1:
The patent extracts only the essential feature learning functionality from conventional neural networks and combines it with a simplified custom classification network. This extraction eliminates unnecessary complexity while preserving the core capabilities needed for accurate identification.
Solution Approach 2:
The patent changes the parameters of the neural network by using a smaller, task-specific classification network with fewer parameters than conventional systems. This reduction in parameters simplifies the system while maintaining accuracy through the powerful feature extraction provided by the pre-trained network.
4Adaptability or versatility
If conventional machine learning systems are configured for general tasks, then they can identify various object types, but they lack accuracy in performing specific tasks such as identifying particular vehicle models
Solution Approach 1:
The patent applies local quality by creating a specialized custom image classification neural network tailored to specific tasks (e.g., vehicle model identification). This custom network is trained on task-specific data to achieve high accuracy in its domain, while the general feature learning network provides broad adaptability across different object types.
Solution Approach 2:
The patent segments the identification task into two parts: general feature extraction (handled by the pre-trained network) and specific classification (handled by the custom network). This segmentation allows each component to be optimized for its specific function, achieving both versatility and precision.
Data Source
AI summary
The system and method are configured to identify cars and obtain information about the identified cars. The system and method involves a software application installed on a computing device that includes a camera configured to capture images. After the software application is installed, the user can activate the camera through the software application and point the camera at a vehicle. The software application includes an object recognition function that can recognize the vehicle in the view of the camera. After recognizing the vehicle, the software application displays information about vehicle on a screen of the computing device. The software application can be used with an event such as an auto show. In addition to identifying the cars, the identified information can be used for other purposes such as a part of a scavenger hunt or other contests. Awards can be given to participants who complete the contest.