1. Introduction
Iron ore is one of the main mineral resources to support the development of the national economy, is the material basis for the development of the iron and steel industry, due to the shortage of mineral resources and non-renewable, how to achieve reasonable, efficient and accurate mining of iron ore resources has become an urgent problem to be solved. Open-pit mining is a large-scale excavation project, and the determination of the distribution of rocks and ore bodies is the first step in open-pit mine production planning, so it is particularly important to clarify the distribution of minerals and rocks in the open-pit stope and accurately extract this kind of information for subsequent mining planning. 
However, traditional ore and rock identification methods (such as mine sampling, chemical testing, etc.) consume large manpower and material resources and are not efficient, and can no longer meet the production needs of open-pit stopes. Hyperspectral remote sensing can achieve a spectral resolution of nanometers, which can capture the diagnostic spectral characteristics of different rocks and minerals, making it possible to use macroscopic technology (hyperspectral remote sensing) to detect microscopic information (rocks and minerals). At present, the extraction and quantitative inversion of rock and mineral distribution information is an important direction of hyperspectral remote sensing geological application, and great progress has been made in extracting alteration mineral information by using mineral spectral mapping, principal component transformation, and band ratio based on spectral feature extraction. The development of imaging spectroscopy technology has greatly promoted the improvement of geological prospecting methods and improved the accuracy of identification. However, the existence of mixed pixels is the main reason why the traditional rock and mineral distribution identification methods are difficult to meet the practical requirements. Hybrid pixel decomposition technology is an important means to interpret pixels, which can decompose the required rock and mineral information from the mixed pixels, and realize the accurate identification and quantitative inversion of rocks and minerals. 
At present, with the continuous emergence and rapid development of new technologies such as modern information technology, artificial intelligence, and big data, traditional mines are moving towards unmanned mining, intelligent mining and intelligent mining. In this study, the open-pit stope was selected as the test site to study the open-pit iron ore detection technology based on UAV imaging spectroscopy, which provided a new technology for the distribution and boundary determination of ore and rock in open-pit mining, and also provided a new idea for the intelligent perception of ore and rock in open-pit unmanned mining.
2. Field test
2.1 Overview of the study area
The dumbaling open-pit stope is an important production area of ore raw materials, located in Anshan City, Liaoning Province. The ore type is a typical Anshan-style iron ore, mainly hematite, mostly poor iron ore, the iron grade is mostly in the range of 20% and 35%, and the surrounding rock mainly includes phyllite, chlorite schist, amphibole, mixed granite and mica schist.
2.2. Data collection and preprocessing
At present, most of the research data of remote sensing geology mainly come from satellite platforms. In recent years, UAV hyperspectral technology has gradually matured and improved, and compared with other sensor platforms, UAV is simple to operate, can adjust the route and flight altitude, and has convenient data acquisition methods. In addition, due to the low altitude and high spatial resolution of UAV remote sensing, the accuracy of ground object recognition is greatly improved, which is suitable for ground investigation in small mining areas.
And due to the short data acquisition cycle, the flexibility and timeliness of stope monitoring data acquisition are guaranteed. Therefore, in this study, the UAV equipped with a hyperspectral imager was used to collect stope remote sensing data, and a series of pre-processing such as atmospheric correction, image correction and image stitching were carried out on the hyperspectral images collected by the UAV. Due to the serious noise in the 900 and 1000nm bands, which masks the spectral information of the ground objects themselves, the spectral information of this band is eliminated in this study and 0Hyperspectral images with 5 m spatial resolution and 245 bands, as shown in Figure 1.
Figure 1 Spectral data of the study area acquired by the UAV
The entire image was cropped to obtain the target study area as shown in Figure 2. The white line in Figure 2 is the distribution boundary of the ore body marked by the mine geology department according to the laboratory test results, the white coil demarcates the area as the ore body, and the surrounding area is the surrounding rock.
Figure 2: Distribution of ore bodies in the study area
2.3. Ore and rock remote sensing identification and ore body demarcation
In this study, the mixed pixel decomposition method was used to identify and extract the distribution of ore and rock in the open-pit stope. Due to the large number of mineral types and complex mineral types in the ore body and surrounding rock, and the small particle size, the use of mineral species as end members is more complicated. The ore type in the ore body is mainly hematite, with a relatively concentrated grade (about 30%), and the surrounding rock near the ore is mainly phyllite. Therefore, in order to simplify the process, hematite is used as the end element in the ore body, and phyllite is used as the end element of the surrounding rock, in addition to a small amount of stagnant water, vehicles and shadow features in the mining area. Finally, five kinds of features were selected as the end members for mixed pixel decomposition, which were hematite, phyllite, ponding, vehicle and shadow. The method of manually selecting the ROI of the region of interest is accurate, reliable, simple and fast, but it requires the operator to have a high level of understanding of the study area.
Fig.3. Spectral curves of each endmember
The spectral curves of the five end members are shown in Figure 3, and according to the comparison, it can be found that the spectral curves of the surrounding rock and the ore body have certain differences, and the phyllite has no obvious spectral characteristics in the 400 900nm band, and the reflectance values are all low, and the distribution is concentrated, mostly at 010~0.15 intervals. Due to the electron jump of Fe3+, the spectra of the ore body form a weak reflection peak feature around 750 nm (red band). The overall reflectivity of the vehicle is higher, the reflectivity of shadows is lower and gentle, and the water absorption is less and the reflectance is lower before the visible band of 600 nm, reaching a peak at 700 nm. Therefore, the hybrid pixel decomposition technique can effectively distinguish different endmembers based on their spectral differences. In this study, the fully constrained least squares algorithm is programmed to operate, and the images of the study area and five end-member spectra are used as input data, and the test results are shown in Figure 4, which are the abundance maps of ore body, surrounding rock, shadow, water accumulation, and vehicle.
Comparing Fig. 4(a) with Fig. 2, it is found that the ore body extracted by the mixed pixel decomposition technique is basically consistent with the ore body distribution delineated by laboratory testing. However, the artificially delineated ore-rock boundaries in Figure 2 are distinct, while the ore-body boundaries delineated by mixed pixel decomposition technology are blurred, which may reflect the real ore-body distribution to a certain extent. In order to quantitatively evaluate the distribution accuracy of the extracted ore body, the results of the two methods were compared. Based on the principle of linear mixed model, the abundance information corresponding to the ore body is used to calculate its distribution area in the open-pit stope according to the following formula.
Fig.4 Abundance distribution of each endmember
where xi is the abundance of the mineral in the ith pixel; r is the spatial resolution m of the image; n is the total number of cells; s is the distribution area of a certain mineral in the stope, m2. After equation (7), it is calculated that the distribution area of the ore body extracted from the experiment in this study is 6236547m2, the delineated area of the ore body in Figure 3 is 6785925m2, the relative accuracy of the area is 9190%, the extraction results are ideal.
3. Summary of this chapter
1) At present, the stope ore and rock identification method is mainly based on the traditional laboratory method, which has shortcomings such as low sampling density, long test period, low efficiency, and lagging grade test results, which leads to inaccurate delineation of ore body boundaries and seriously affects subsequent production. The hybrid pixel decomposition technology was used to collect the imaging spectral data of the study area by using the unmanned aerial vehicle (UAV), and the identification and automatic extraction method of ore and rock in the open-pit iron ore stope were studied. The field test results show that the open-pit iron ore delineation technology based on UAV hyperspectral technology can effectively delineate the iron ore body, and the accuracy is higher than that of the existing indoor laboratory delineation ore body area, which provides a new method for realizing the intelligent perception of ore and rock.
2) Based on hyperspectral remote sensing technology, only the ore and rock identification analysis was carried out in this study, and the ore grade inversion study will be carried out in the future.
Recommended:
UAV airborne hyperspectral imaging system iSPEChyper-VM100
A hyperspectral imaging system based on a small multi-rotor UAV is composed of a hyperspectral imaging camera, a stabilized gimbal, an airborne control and data acquisition module, and an airborne power supply module. The UAV airborne hyperspectral imaging system effectively solves the problem of poor image quality caused by vibration when the micro UAV is equipped with a pushbroom hyperspectral camera through unique built-in or external scanning and stabilization control, and has high spectral resolution and good imaging performance.