Cascade Neural Network for Target Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target detection methods face challenges in improving detection rates due to interference from factors like face posture, skin color, dirt, illumination, occlusion, and blurriness, and are limited by large storage requirements and slow calculation speeds, making them unsuitable for low-specification hardware devices.
Innovation Solution
A target detection method using a cascade neural network with multiple parallel sub-neural networks for classifying candidate areas, reducing the number of neural networks and storage requirements, and improving calculation speed by generating an image pyramid and fine-tuning neural networks for enhanced detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If depth-learning based target detection method is used, then detection rate is improved, but storage requirement and calculation complexity increase
Solution Approach 1:
The patent segments the deep learning classification model into multiple components: image pyramid generation, candidate area extraction, and classification using trained neural networks. This segmentation allows selective processing and reduces the storage burden by only retaining essential model parameters rather than entire large-scale models.
Solution Approach 2:
The patent extracts only the critical classification functionality from the full deep learning model, isolating the neural network parameters needed for classification while discarding or compressing redundant components. This extraction enables using deep learning detection capabilities with reduced storage requirements.
2Reliability
If depth-learning based target detection method is used, then detection rate is improved, but calculation speed decreases
Solution Approach 1:
The patent performs preliminary actions by generating the image pyramid and extracting candidate areas before applying the classification model. This preprocessing organizes the data in advance, allowing the neural network to process only relevant regions rather than the entire image, thereby improving calculation speed while maintaining detection accuracy.
Solution Approach 2:
The patent applies partial action by focusing computational resources only on candidate areas that are likely to contain targets, rather than processing the entire image. This selective processing maintains high detection rates while significantly reducing the computational burden and improving speed.
3Measurement precision
If large quantity classification model is used, then detection accuracy is improved, but processor resource consumption increases
Solution Approach 1:
The patent applies local quality by training different neural networks for different target attributes (e.g., face detection, object detection) and applying only the relevant networks to specific detection tasks. This localized approach maintains high detection accuracy for each specific task while reducing overall processor resource consumption by avoiding unnecessary computations.
4Reliability
If depth-learning based method is used in low-performance equipment, then detection capability is improved, but hardware requirements are not met
Solution Approach 1:
The patent employs lightweight neural network models that can be deployed on low-performance hardware. By using simplified network architectures and selective preprocessing, the system achieves acceptable detection capability on resource-constrained devices without requiring high-end hardware specifications.
Data Source
AI summary
A method of detecting a target includes generating an image pyramid based on an image on which a detection is to be performed; classifying candidate areas in the image pyramid using a cascade neural network; and determining a target area corresponding to a target included in the image based on the plurality of candidate areas, wherein the cascade neural network includes a plurality of neural networks, and at least one neural network among the neural networks includes parallel sub-neural networks.