Modular Adaptive System Based on a Multi-Stage Neural Structure for Recognition of 2D Objects of Discontinuous Production

TL;DR

This paper introduces a multi-stage neural system that achieves invariant recognition of 2D objects with overlapping classes, using DCT spectrum transformation and neural networks for filtering, recognition, and classification, suitable for real-time applications.

Contribution

A novel multi-stage neural architecture utilizing DCT spectrum features for invariant 2D object recognition in discontinuous production environments.

Findings

High accuracy and flexibility demonstrated in tests with 2D objects.

Effective partial and full rotation invariance achieved.

System offers fast processing and easy reconfiguration for real-time use.

Abstract

This is a presentation of a new system for invariant recognition of 2D objects with overlapping classes, that can not be effectively recognized with the traditional methods. The translation, scale and partial rotation invariant contour object description is transformed in a DCT spectrum space. The obtained frequency spectrums are decomposed into frequency bands in order to feed different BPG neural nets (NNs). The NNs are structured in three stages - filtering and full rotation invariance; partial recognition; general classification. The designed multi-stage BPG Neural Structure shows very good accuracy and flexibility when tested with 2D objects used in the discontinuous production. The reached speed and the opportunuty for an easy restructuring and reprogramming of the system makes it suitable for application in different applied systems for real time work.