Abstract: The recognition and classification of mechanical components is a key process on the
manufacturing automation line. In terms of the classification of the mixed and cleaned engine main
bearing cap, through the analysis of the actual characteristics of the main bearing cap parts, the
classification and recognition method for the main bearing cap based on SIFT-SVM was proposed.
The method first extracted all scale-invariant feature transform (SIFT) feature vectors of the training
dataset image, then employed the K-means clustering method to cluster all feature vectors into K
classifications, and substituted the obtained K clustering results into the bag of word model (BoW).
“Vocabulary” was utilized to describe each training image, thereby obtaining a BoW description of
the image. The BoW description of each image served as a training input, and the classification model
for the main bearing cap was trained using a support vector machine (SVM). The experimental results
show that under the calibrated lighting conditions, the recognition rate of the main bearing cap parts
can reach 100%, and the recognition time for a single part was 0.6 seconds, which verified the
effectiveness and efficiency of the algorithm.

Key words: parts recognition and classification, machine vision, scale-invariant feature transform;
word bag model, support vector machine classifier