Convolutional Neural Network Segmentation for Object Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis technologies using artificial neural networks often perform unnecessary operations across multiple image frames, leading to increased processing speed and energy consumption, necessitating a method to reduce these costs by considering object features.
Innovation Solution
The method involves dividing a convolutional neural network into groups based on object occurrence frequency and importance, where each group processes specific regions of interest in images, using only the necessary convolution layers to determine object classes, thereby optimizing processing speed and energy consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a convolutional neural network performs uniform operations on all image frames, then object recognition accuracy is maintained, but processing speed decreases and energy consumption increases
Solution Approach 1:
The patent segments the convolutional neural network into multiple groups of convolution layers, where each group is responsible for detecting specific object classes. This segmentation allows the system to process only relevant features for each object type rather than performing uniform operations on all image frames, thereby improving processing speed while maintaining recognition accuracy.
Solution Approach 2:
The patent applies local quality by assigning different processing depths to different object classes based on their importance and occurrence frequency. Critical objects are processed with deeper convolution layer groups for higher accuracy, while less important objects use shallower processing, optimizing the balance between speed and accuracy for each local region of interest.
2Reliability
If a convolutional neural network performs uniform operations on all image frames, then comprehensive object detection is achieved, but energy consumption increases
Solution Approach 1:
The CNN is divided into multiple convolution layer groups, each specialized for detecting specific object classes. This segmentation enables the system to activate only the necessary groups for each image frame based on the presence and importance of objects, reducing overall energy consumption while maintaining comprehensive detection capabilities.
Solution Approach 2:
The system performs partial action by processing only the necessary convolution layer groups required for detecting objects in each specific image frame. Instead of always executing the full network depth, it applies just enough processing to achieve reliable detection, reducing energy consumption while maintaining detection comprehensiveness.
3Measurement precision
If the convolutional neural network processes all convolution layers for every object, then detection accuracy is maximized, but processing complexity increases
Solution Approach 1:
The patent segments the convolutional layers into multiple groups organized by object class importance and occurrence frequency. This segmentation simplifies the processing complexity by allowing the system to selectively activate only the necessary groups for each detection task, rather than managing a monolithic complex network structure.
Solution Approach 2:
The system dynamically selects which convolution layer groups to activate based on the specific detection requirements of each image frame. This dynamic approach reduces processing complexity by adapting the network depth and structure to the actual content being processed, rather than using a fixed complex structure for all cases.
4Productivity
If the system focuses on critical objects only, then processing speed increases, but object detection coverage decreases
Solution Approach 1:
The patent segments objects into different priority levels and assigns appropriate convolution layer groups to each level. This segmentation enables the system to maintain high processing speed for critical objects while still providing detection coverage for other objects through shallower processing groups, thus balancing speed and coverage.
Solution Approach 2:
The convolution layer groups are designed with multi-functionality, where higher-level groups can process both critical and non-critical objects, while lower-level groups specialize in critical objects. This universal design allows the system to maintain comprehensive detection coverage while optimizing processing speed for priority objects.
Data Source
AI summary
Provided are a method and electronic device for inferring an object in an image using a convolutional neural network (CNN) model. The method includes including: identifying a first region of interest in a first frame in the moving image, and a first object in the first region of interest, by providing the first frame to convolution layer groups sequentially connected in the CNN model, identifying a second region of interest in a second frame, the second region of interest corresponding to the first region of interest, and the second frame being after the first frame, providing the second region of interest to the CNN model, and obtaining first output data output from a first convolution layer group from among the convolution layer groups, and determining whether to identify a second object in the second region of interest by using a second convolution layer group, based on the first output data.