Spraying object increment identification system and method integrating multiple modules and spiking neural network
By integrating attention mechanisms, memory gating networks, and task memory matrices into a spiking neural network, the catastrophic forgetting problem of spill identification models in open traffic scenarios is solved, achieving efficient identification and continuous learning of new and old spill categories.
CN122135313APending Publication Date: 2026-06-02JINLING INST OF TECH
7 Cites 0 Cited by
Patent Information
- Application Number
- CN202610235022.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-27
- Publication Date
- 2026-06-02
AI Technical Summary
Technical Problem
Existing debris recognition models struggle to adapt to complex and dynamic environments in open traffic scenarios, suffer from catastrophic forgetting problems, consume high energy, and are unable to quickly identify new types of debris.
Method used
A spiking neural network integrating an attention mechanism, a memory gating network, a task memory matrix, and a feature fusion module is used to protect old task knowledge through orthogonal projection, enabling incremental learning of the types of projectiles.
Benefits of technology
It improves the accuracy and anti-forgetting ability of spilled material recognition, supports continuous learning in dynamic scenarios, and achieves stable recognition of new and old spilled material types.
✦ Generated by Eureka AI based on patent content.
Abstract
The invention discloses a throwing object increment recognition system and method integrating multiple modules and a pulse neural network. The system comprises a pulse feature extraction module, an orthogonal projection module, an attention mechanism module, a task memory matrix, a memory gating network, a feature fusion module and a classifier module. The method comprises the following steps: carrying out pulse coding on an input throwing object image, and extracting pulse features; pulse features are processed through an orthogonal projection module, and learned task knowledge is protected; inputting the pulse characteristics into an attention mechanism module and a memory gating network in sequence, and performing knowledge retrieval in combination with a task memory matrix; output features from the attention mechanism module, the memory gating network and the task memory matrix are fused in a feature fusion module and input into a classifier module, so that a throwing object category identification result is obtained; when a new task is learned, the lateral connection weight of the orthogonal projection module, the attention mechanism module, the task memory matrix and the memory gating network parameters are updated, and related parameters of an old task are frozen at the same time.
Need to check novelty before this filing date? Find Prior Art