Accurate identification and continuous extraction of fissures in loess areas based on unmanned aerial vehicle visible light images.

In view of the low accuracy and incompleteness of surface fissure extraction in mining areas, this paper takes the surface fissures in the visible light image of a mine in western China as an example, enhances the difference of spectral characteristics of the image by hue–saturation–intensity transformation, and then gets the rough extraction of ground fissures by combining the edge detection algorithm based on Sobel operator and Prewitt operator. Aiming at the discontinuity of ground fissures in remote sensing images due to the weak contrast with the environmental background, this paper uses the regional growth method based on pixel and object-oriented to realize the continuous expression of fracture areas. First, the growth area of ground fissure seed points is established, and the similarity judgment criterion based on the spectral characteristics and the real morphological characteristics of the fissure are proposed. Then, the pixels meeting the conditions are merged with the extracted objects to solve the problem that which is difficult to continuously extract ground fissure under the influence of "same object different spectrum" and vegetation coverage. The results of uniform sampling analysis show that the extraction integrity of fissures with width greater than 20 cm in 2.1 cm/pixel visible light image is 98.69%. The extraction integrity of fissures with width 10–20 cm is 86.54%. Part of fissures with width < 10 cm were also extracted, and the extraction integrity was 58.62%. The average extraction accuracy of fissures width fluctuates around 80%. [ABSTRACT FROM AUTHOR]