Convolutional Neural Network Pixel-Level Heat Map Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional convolutional neural networks sacrifice resolution accuracy and the ability to identify multiple objects within an image due to their convolutional and combinatory nature, making them less effective for precise object location detection.
Innovation Solution
A knowledge manager processes an image using a convolutional neural network to generate a pixel-level heat map, analyzing decision points corresponding to target objects and marking regions indicating their locations, thereby improving object detection accuracy and capability to identify multiple objects within an image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional convolutional neural networks with fully connected layers are used to detect object locations, then general object detection capability is achieved, but resolution accuracy and the ability to identify multiple objects deteriorate
Solution Approach 1:
The patent removes the fully connected layers from the traditional convolutional neural network architecture, extracting only the essential convolutional components needed for location detection. This extraction eliminates the combinatorial operations that cause information loss while preserving the network's ability to detect objects, thereby improving resolution accuracy without the detrimental effects of full combinatorial processing.
Solution Approach 2:
The patent segments the object detection task into independent pixel-level classification decisions. Instead of using fully connected layers that combine all pixel information into a single detection decision, the network processes each pixel independently through convolutional operations, allowing multiple objects to be identified simultaneously while maintaining high resolution accuracy for each detected location.
2Adaptability or versatility
If traditional convolutional neural networks combine information across the entire image, then general detection capability is achieved, but the ability to identify multiple objects deteriorates
Solution Approach 1:
The patent extracts and eliminates the fully connected layers that perform combinatorial information processing across the entire image. By removing this information-combining step, the network preserves spatial information at pixel level, enabling it to identify multiple distinct objects simultaneously without the information loss that occurs when all pixel data is aggregated into a single detection decision.
Solution Approach 2:
The patent changes the output dimensionality from a single combined detection result to a pixel-level heat map where each pixel contains independent classification information. This dimensional transformation allows the network to maintain adaptability for detecting multiple objects across different locations while avoiding information loss by preserving the spatial distribution of detection confidence at each pixel position.
Data Source
AI summary
An approach is provided in which a knowledge manager processes an image using a convolutional neural network. The knowledge manager generates a pixel-level heat map of the image that includes multiple decision points corresponding to multiple pixels of the image. The knowledge manager analyzes the pixel-level heat map and detects sets of decision points that correspond to target objects. In turn, the knowledge manager marks regions of the heat map corresponding to the detected sets of per-pixel decision points, each of the regions indicating a location of the target objects.