This paper is the result of geological hazard survey engineering.

Objective The loess landslide is a major hidden danger to the safety of human settlements and urban construction in the loess region. Landslide identification is the foundation of other research work on landslide disasters. By utilizing unmanned aerial vehicle oblique photogrammetry three-dimensional (3D) modeling, loess landslides can be intuitively and quickly identified from different dimensions and perspectives, enabling the extraction of feature parameters. This can provide technical support for risk identification and refined risk management research of loess landslides.

Methods  While researching Heiniugou Village in the southwest of Honghe Town, Pengyang County, Guyuan City, and the Ningxia Hui Autonomous Region, regional loess landslide identification and feature parameter extraction and analysis were carried out using unmanned aerial vehicle oblique photogrammetry data acquisition, 3D modeling, on-site verification, and geostatistical analysis.

Results  Based on a 3D−real−life model, we located steep walls and their surrounding shapes in the geomorphic depression areas along the valley in the study area. By combining color tone, texture, and micro-geomorphology indicators, we were able to identify all of the loess landslides in the specified region. A total of 23 landslides were delineated, and two non-landslide points were removed through on-site verification. The remaining 21 landslides were densely distributed at the mouth of the main and branch gullies, appearing to slide towards each other from opposite sides of the gullies and exhibiting a mass occurrence. The proportion of large and super−large landslides reached 57.14%. The landslides primarily slid to the southwest (sunny slope) and to the southeast (semi-sunny slope), with relative height differences between 80–120 m. The slopes of these landslides were mostly concave and measured between 20°–30°. The sites of these landslides were mainly sources of vegetation, bare land, or farmland, with a small percentage of the land made up of roads and rivers.

Conclusions A 3D-realistic model based on unmanned aerial vehicle oblique photogrammetry can accurately and quickly identify regional loess landslides from multiple dimensions and perspectives and analyze their related feature parameters, which can make up for the shortcomings of current two-dimensional planar remote sensing images. Moreover, this process can also provide data support for research on landslide susceptibility, danger, vulnerability, and risk assessment.