Mechanical Part Recognition based on Moment Invariant and BP Neural Network

• Haitao Wang
• Weijian Mao
• Mengyuan Xu
• Hui Xu

This paper solves the problem of mechanical parts identification based on MTALAB software using the moment invariants theory first proposed by Hu. M.K. In order to simulate the identification of mechanical parts, we added different degrees of noise to the randomly selected three mechanical parts pictures. Firstly, the images of mechanical parts are processed by median filtering, binarization and Canny operator edge detection algorithm. Then, the images are rotated at equal intervals and scaled at different multiples. Some images are selected from the rotated and scaled images as training sample images. The number of input and output nodes and the number of hidden layer neurons of BP neural network are determined by the moment invariants of samples and the number of selected mechanical parts. The 7-13-3 BP neural network structure with a single hidden layer was established to identify the selected mechanical parts. Using the Levenberg-Marquardt training algorithm, the BP neural network is trained with the moment invariants of the training sample images. The minimum training error is 0.001, the learning rate is 0.01, and epochs are 3000. Using the remaining images and noise-processed images as the test sample of the neural network, the test accuracy rate is close to 90%, solving the problem of mechanical parts identification

Select Volume / Issue:

• Volume 6,Issue 2,Mar.,2019

Year:

2019

Type of Publication:

Article

Keywords:

BP Neural Network; Image Preprocessing; Mechanical Part Identification; Moment Invariant

Journal:

IJASM

Volume:

6

Number:

2

Pages:

23-30

Month:

March

ISSN:

2394-2894