Infrared image-based identification method for the gradation of rock grains using heating characteristics

Abstract

In earth-rock dam engineering, the gradation characteristics of filling materials have an important influence on the compactability, impermeability, and stress–strain characteristics of dams. A sound gradation is the key to guaranteeing dam safety and stability. Based on the principles of the heat conduction and heating characteristic tests on rocks, the power function equation between volume and temperature increase for pebble and gravel grains in a heating environment is established with correlation coefficients of 0.98. Aiming at the low span of gray-scale change of digital infrared images, the linear gray-scale transformation method with a Gamma factor of 1 to enhancing the digital infrared images, and the maximum between-cluster variance method (OSTU method) is adopted to perform binary grain target-context segmentation on enhanced images. With regard to the phenomenon of grain adhesion, it subjects flattened images after binary segmentation to peak processing using Dis operation, and introduces an improved watershed algorithm (i.e., local minimum with threshold) for image reconstruction, effectively segmenting adhesive grains while avoiding over-segmentation. According to maximum stability theory, the shortest side of the minimum enclosing rectangle (MER) of the grains in images is proposed as the judgment index for grain sieving. Five gradation detections were carried out on pebbles and gravels, respectively. For grain mass fraction identification results for various grain size intervals, the maximum absolute errors for pebbles and gravels do not exceed 6% and 7%, and the average absolute errors do not exceed 4% and 5%, respectively. Thus, the identified gradation curve fits well with the mechanical sieving gradation curve, and can meet engineering application requirements. This new method overcomes several deficiencies of traditional methods for gradation image detection, such as the susceptibility of image sampling to various interfering factors and difficulties in controlling errors in calculating grain mass, and therefore, provides a novel conceptual technology for particle grading image recognition.