Object Detection Model Using Local and Non-Local Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection methods in satellite images face challenges in accurately identifying objects obscured by clouds or other obstructions, as they primarily focus on relationships between adjacent pixels, neglecting the importance of non-adjacent pixel relationships.
Innovation Solution
The method employs an object detection model that combines local blocks for computing relationships between adjacent pixels and non-local blocks for computing relationships between non-adjacent pixels, using weighted sum computations and residual connections to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only local blocks are used to compute relationships between adjacent pixels, then the computational process is simple and fast, but the accuracy of object detection deteriorates when objects are obscured by clouds
Solution Approach 1:
The patent merges local blocks and non-local blocks into a single object detection model. The local block processes adjacent pixel relationships for basic feature extraction, while the non-local block processes non-adjacent pixel relationships to capture global context and infer obscured objects. This combination resolves the contradiction by integrating both simple local processing and complex global reasoning to achieve high detection accuracy without excessive complexity.
Solution Approach 2:
The patent segments the object detection model into distinct local and non-local blocks, each handling specific computational tasks. The local block handles adjacent pixel computations for immediate feature detection, while the non-local block handles non-adjacent pixel computations for contextual inference. This segmentation allows the system to manage complexity through modular design while achieving comprehensive detection accuracy.
2Reliability
If non-local blocks are added to compute relationships between non-adjacent pixels, then the accuracy of detecting obscured objects improves, but the computational time and processing complexity increase
Solution Approach 1:
The patent combines local and non-local blocks in a coordinated manner where the local block performs rapid adjacent pixel processing first, followed by the non-local block for non-adjacent pixel processing. This merging strategy ensures that reliable detection of obscured objects is achieved through complementary processing, with the local block providing quick basic detection and the non-local block providing accurate contextual inference.
Solution Approach 2:
The local block performs preliminary computation on adjacent pixels to extract basic features and identify obvious objects before the non-local block processes non-adjacent pixels for contextual inference. This preliminary action reduces the computational burden on the non-local block and optimizes the overall processing time while maintaining high reliability for obscured object detection.
Data Source
AI summary
According to an exemplary embodiment of the present disclosure, a method of detecting an object is disclosed. The method of detecting an object includes computing an image including an object by using an object detection model including a local block and a non-local block, in which the local block computes a relationship between adjacent pixels included in a feature map, and the non-local block computes a relationship between non-adjacent pixels included in the feature map.