Research on Intelligent Classification of Aggregates based on SegFormer

Abstract

Aggregate grading refers to the proportional relationship between the different particle sizes that make up an aggregate, and is a vital part of aggregate production. Aggregate grading is the first step in aggregate grading and the grading is of great importance to the quality of the aggregate grading. Traditional methods of grading artificial aggregates include the sieve method and the laser method. These methods are inefficient in performing aggregate classification and the accuracy of the test is subjectively influenced by the inspector. To solve this problem, the SegFormer algorithm was used to achieve intelligent classification of aggregates. The dataset was produced by photographing and labelling 200 top and 200 side views each from samples of aggregates provided by local aggregates companies. The SegFormer algorithm was used to semantically segment the image. Aggregate particle size was obtained by calculating the minimum circumscribed circle diameter of the coloured area in the top view segmentation result and the maximum inscribed circle diameter of the coloured area in the side view segmentation result to obtain the aggregate thickness. Combined aggregate size and thickness to determine the type of aggregate. The experiments show that the combination of SegFormer algorithm to achieve high accuracy and speed of intelligent classification of aggregates is of significant importance to the intelligent development of this industry.