Dynamic Anchor Parameter Adjustment for Target Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target detection methods based on deep learning are limited by a fixed number of anchor types during training, resulting in a low object sample coverage rate and inaccurate detection of specific objects, as they cannot dynamically adjust anchor parameters during practical application.
Innovation Solution
A target detection method that acquires and inputs preset anchor parameters, including anchor scales and aspect ratios, into a target detection model, allowing for dynamic detection and adjustment of network parameters based on external anchor parameters, thereby increasing the types of anchors and improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fixed type of anchor is set only in the training process, then the network parameter can be obtained by training, but the object sample coverage rate is low and detection accuracy for specific objects is poor
Solution Approach 1:
The patent applies the dynamics principle by enabling the target detection model to dynamically adjust anchor parameters (scale and aspect ratio) during the detection process. Instead of using fixed anchor types determined during training, the model can adaptively select and adjust anchor parameters based on the specific image being detected, thereby improving both detection accuracy and object sample coverage rate for various object types and scales
Solution Approach 2:
The patent implements parameter changes by modifying the anchor parameters (scale and aspect ratio) that are input to the target detection model. By changing these parameters dynamically during detection rather than keeping them fixed, the system can better adapt to different object sizes and shapes, resolving the contradiction between detection accuracy and coverage rate
2Adaptability or versatility
If the network parameter is determined after training, then the model structure is simplified, but the model cannot dynamically adjust to detect objects with various aspects
Solution Approach 1:
The patent applies dynamics by introducing dynamic anchor parameter adjustment capability into the target detection model. The model can now adaptively modify anchor scales and aspect ratios during detection based on input images, enabling it to handle objects with various aspects without requiring a completely complex retraining process for each scenario
Solution Approach 2:
The patent implements universality by designing a target detection model that can handle multiple object types, scales, and aspects using a single unified framework. By accepting anchor parameters as input and dynamically adjusting them, the model achieves multi-functionality in detecting diverse objects without needing separate specialized models for each object category
Data Source
AI summary
The present disclosure provides a target detection method, target detection system and non-volatile storage medium. The target detection method includes: acquiring an image to be detected and anchor parameters of preset type number, wherein the anchor parameters are parameters of an anchor set on the image to be detected, and each type of anchor parameters include an anchor scale and an anchor aspect ratio; inputting the image to be detected and the anchor parameters into a target detection model; and carrying out target detection on the image to be detected on the basis of the anchor parameters by the target detection model to obtain a detection result, the detection result including a category and/or a position of a target object included in the image to be detected.