黑土地是地球上极为珍贵的土壤资源，其土壤性状好、有机质含量高，被誉为“耕地中的大熊猫”。然而，由于长期高强度的利用加之土壤侵蚀，造成了黑土层流失、土壤退化等问题，黑土层逐渐变薄、变瘦、变硬。

东北黑土地是我国“黄金玉米带”分布区，也是我国主要的玉米种植区。东北黑土地玉米生长季为4个月左右，强降水集中易导致水土流失，玉米植被覆盖度是表征黑土地利用程度、水土流失及土层变薄风险的指示器；黑土区在较长的非生长季无耕期内风蚀风险高，玉米秸秆覆盖监测是量化黑土地防风固土水平和保护性耕作实施程度的重要措施。因此，对我国黑土地玉米种植区实施作物全年覆盖监测，即生长季的植被覆盖监测与非生长季的秸秆覆盖监测，对黑土地利用与保护具有非常重要的意义。

国产高分系列卫星影像空间分辨率高、重访周期短、观测幅宽大，在监测黑土地全年覆盖方面具有很大的优势。但是，目前利用高分系列卫星影像进行黑土地玉米种植区全年覆盖监测方面的研究较少，这不利于黑土地玉米种植区全年覆盖动态、快速监测能力的提升，也影响了我国自主遥感影像数据应用价值的有效发挥和应用潜力的科学评估，制约了我国自主遥感卫星应用的进一步发展。

针对国产宽幅多光谱卫星影像因观测角度不一致造成辐射差异明显、作物植被/秸秆覆盖度监测成果因土壤背景差异显著产生较大误差、玉米秸秆高覆盖地块识别精度较低等问题，本文选择黑土地保护性耕作“梨树模式”创建地吉林省梨树县作为研究区，以国产高分系列卫星影像为数据源，探索了黑土地玉米种植区全年覆盖监测方法：研究了GF-1 WFV多光谱影像的辐射一致性校正方法，以解决影像幅宽大、倾斜观测带来的辐射差异问题；构建了考虑黑土地土壤背景差异的像元二分法模型，并基于GF-1 WFV和GF-6 WFV影像分别估算了作物植被覆盖度和秸秆覆盖度；基于GF-2 PMS和GF-1 B/D PMS高分辨率影像，研究了基于深度学习U-Net模型的玉米秸秆高覆盖地块识别方法；在梨树县玉米种植区开展了全年覆盖监测方法验证。主要研究成果如下：

（1）研究了GF-1 WFV多光谱影像的辐射一致性校正方法，以减轻国产高分影像因观测角度不一致带来的影像辐射差异。

针对国产高分卫星影像用于植被参数定量估算时，面临的因幅宽大、观测角度不一致造成的辐射差异问题，本文研究了基于MODIS BRDF产品MCD43的GF-1 WFV多光谱影像的辐射一致性校正方法。研究结果表明：经过辐射一致性校正，GF-1 WFV NDVI与Sentinel-2 MSI NDVI的相关性得到提升，决定系数R²由0.793增加至0.805，偏差则由-0.043下降至-0.017，均方根误差由0.074降至0.061。这说明经过辐射一致性校正，可减轻GF-1 WFV影像因倾斜观测带来的辐射差异。

（2）研究了基于黑土地土壤质地分区的像元二分法模型，探索了考虑土壤背景差异的玉米植被覆盖度估算方法；挖掘GF-6 WFV影像新增光谱波段信息，构建了秸秆光谱指数，进而探索了考虑土壤背景差异的玉米秸秆覆盖度估算方法。

基于黑土地土壤表层（0-5cm）砂粒含量，分区构建了考虑土壤背景差异的基于像元二分法的植被覆盖度估算模型；对比分析采用土壤质地分区建模前后的植被覆盖度估算精度，决定系数R²由0.888提高到0.973，均方根误差由0.049降低到0.023，偏差由0.021降低至0.010，证明了采用土壤质地分区建模法能够提升植被覆盖度估算精度。

挖掘GF-6 WFV影像新增波段的光谱特征，本文构建并优选出5种秸秆光谱指数；基于最优指数NDI47的秸秆覆盖度估算结果，与基于Sentinel-2 MSI、Landsat8 OLI影像最优指数估算结果的决定系数R²分别为0.847、0.752，表明GF-6 WFV影像在秸秆覆盖度估算方面具有很高的应用可靠性；对比分析采用土壤质地分区建模前后的秸秆覆盖度估算精度，决定系数R²由0.769提高到0.822，均方根误差由0.146降低到0.115，这表明采用土壤质地分区建模法能够提升秸秆覆盖度估算精度。

（3）构建了基于U-Net卷积神经网络的玉米秸秆高覆盖地块识别方法，实现了基于高空间分辨率影像的玉米秸秆高覆盖地块精细识别。

基于GF-2 PMS影像构建了秸秆识别学习标签并进行深度学习网络训练，进而实现了玉米秸秆高覆盖地块识别。研究结果表明：基于U-Net卷积神经网络的玉米秸秆高覆盖地块识别总体分类精度为89.29%，Kappa系数为0.785，用户精度和制图精度为94.18%和93.12%，识别精度高。采用迁移学习方式将秸秆高覆盖地块识别模型迁移至新数据源GF-1 B/D PMS影像，从而实现整个梨树县范围内的玉米种植区秸秆高覆盖地块识别，识别结果制图精度为87.1%。这说明利用U-Net深度学习法能够有效挖掘高空间分辨率遥感影像上玉米秸秆覆盖的光谱特征、纹理特征及结构特征，有助于解决因类型多样、影像特征复杂导致的秸秆覆盖识别精度低的问题。

（4）提出了基于国产高分系列卫星影像的黑土地玉米种植区全年覆盖遥感监测方法，并验证了其应用效果。

针对国产高分系列卫星的载荷特征，结合东北玉米种植区一年一熟耕作制度和土壤异质性显著的特点，本文提出了基于土壤质地分区的黑土地玉米种植区全年覆盖监测方法，并以GF-1 WFV、GF-1 B/D PMS、GF-2 PMS、GF-6 WFV国产高分系列影像为数据源，在黑土地保护“梨树模式”创建区梨树县，开展了玉米种植区全年覆盖监测与验证分析。

研究结果表明梨树县玉米种植区全年保持在较高的覆盖水平：2020年5月下旬-9月下旬为玉米生长季，其植被覆盖度在6月中旬前较低（<0.15），6月下旬至9月上旬，植被覆盖度处于>0.6的高水平；直至收获期又降低至0.15以下，这是植被覆盖对黑土地的利用及保护情况。2020年1月至4月、10月中旬至年底为玉米非生长季，在此期间平均秸秆覆盖度超过0.4，高于保护性耕作要求（秸秆覆盖度≥0.3）；11月上旬，135 825公顷玉米地为秸秆高覆盖地块，这是秸秆覆盖对黑土地的保护情况。

基于国产高分系列卫星影像的玉米种植区全年覆盖监测结果表明，利用本文提出的全年覆盖监测方法，能够利用国产高分系列卫星影像空间分辨率高、重访周期短、观测幅宽大的优势，实现黑土地玉米种植区全年覆盖的精细、快速、动态监测，为挖掘高分卫星载荷在黑土地保护实施监测方面的潜力提供科学依据，具有推广价值与应用指导意义。

The black soil is one kind of precious natural resources, with good soil properties and high organic matter content, which is perfect for planting crops with high yield and good quality. Unfortunately, the black soil in Northeast China has beeb degenerated incrementally resulting from high intensity of utilization coupled with soil erosion. The black soil is becoming thinning and hardening, and the organic matter content is decreased slowly during this degeneration process. Therefore, it is urgent to monitor the utilization of black soil and protect the black soil in order to guarantee the food security and sustainable development of agriculture in our country.

Corn is an important grain and industrial crop in black soil of Northeast China. It is very important to monitor the crop vegetation cover and residue cover all year around, which is an effective way to protect the black soil for keeping the food security in our country. The fractional vegetation coverage in growing season and the residue coverage in non-growing season of corn reveal the utilization and protection of black soil, respectively. Therefore, it is very important to monitor the coverage all year around of corn planted area in black soil. Unfortunately, there is few studies on monitoring the corn vegetation cover in growing season and residue cover in non-growing season all year round for black soil protection currently. Chinese GaoFen series satellites are composed of satellites with different tracking altitudes and different spectral bands from visible bands to microwave bands. Therefore, the earth surface observation can be done using Chinese GaoFen series satellite images. Consequently, the Chinese GaoFen series satellite images are suitable for monitoring the crop vetetation cover and crop residue cover all year around. However, there are few studies on the estimation of crop vegetation cover and crop residue cover in black soil using Chinese GaoFen satellite images currently. Therefore, it is urgent to mining the potentiality of Chinese GaoFen satellite images for monitoring the crop vegetation cover and residue cover aiming at black soil protection, which will promote the further developments and applications of China's autonomous remote sensing satellite.

Herein above statements, this paper is aiming at mining potentiality of the Chinese GaoFen satellite images for monitoring the corn vegetation cover in growing season and residue cover in non-growing season all year round. Firstly, the relative radiation correction of GF-1 WFV multi-spectral image and the developing of fractional vegetation cover estimation model are done to estimate the effective fractional vegetation cover in the growing season of corn. Secondly, the spectral index of crop residue based on the spectral bands of GF-6 WFV multi-spectral image and the crop residue cover estimation model are developed and optimized for the crop residue cover estimation in non-growing season. Lastly, the corn residue high covered area is identified using U-Net deep learning method based on GF-2 PMS and GF-1 B/D high spatial resolution images. The results of these studies are of great significance to black soil protection in Northeast China. The study results are as follows:

 (1) The relative radiometric calibration of Chinese GF-1 WFV multi-spectral satellite images are studied to improve the radiometric consistency.

The relative radiometric calibration method of GF-1 WFV multispectral images is done using the BRDF Bidirectional Reflectance Distribution Function and Albedo (BRDF/Albedo) products of MCD43 for solving the problem of spectral difference caused by inconsistent observation angles of wide scanning of GF-1 WFV scanner. The study resuts indicate that the GF-1 WFV2 NDVI is cloaser to the Sentinel-2 MSI NDVI which is approaching the nadir observation. The coefficients of determination (R²) is improved from 0.793 to 0.805, the RMSE decreases from 0.074 to 0.061, and the bias decreases from -0.043 to -0.017. These results indicate that the relative radiometric calibration method used in this study can calibrate the radiometric inconsistency resulting from inconsistent scanning angle.

(2) The model of pixel dichotomy by considering the background of black soil is proposed. And the vegetation cover in growing season and the residue cover in non-growing season of corn is estimated using the proposed pixel dichotomy model by GF-1 WFV and GF-6 WFV multi-spectral images, respectively.

The content of sand within 0-5 cm surface soil is considered in proposed pixel dichotomy model, which is used to estimate the fractional vegetation cover and residue cover of corn. And the black soil background is considered during this estimation.

The advantages of GF-6 WFV multispectral image lies in the high spatial resolution, short revisit period and wide observation range. Furthermore, there are 3 new spectral bands including red-dege, ultraviolet band and yellow band, in addition to the original 4 spectral bands including blue, green, blue and near infrared bands. These advantages make sure the great potentional for GF-6 WFV multispectral images used in estimating the corn residue cover in black soil in Northeast China. Therefore, the estimating method of corn residue cover in black soil using GF-6 WFV multispectral images is done in this study. There are 10 spectral indexes are developed for estimating the corn residue cover in this study by considering the multispectral characteristics of GF-6 WFV images. And these 5 spectral indexes including NDI87, NDI37, NDI47, NDI32 and NDI38 are found to be more correlated with the measured residue coverage measured in the field. The determination coefficient R² between the estimated and measured residue cover is greater than 0.5, which can explain more than 50% information of corn residue coverage. This result reveals that the emerging purple and yellow bands of GF-6 WFV multispectral images have great potential in corn residue cover estimation in black soil.

By considering the black soil background in Dimidiate Pixel Model, the fractional vegetation cover estimation is improved with determination coefficient R² is improved from 0.888 to 0.973, RMSE is decreased from 0.049 to 0.023, and Bias is decreased from 0.021 to 0.010. For the corn residue cover estimation, the determination coefficient R² is improved to 0.822 and RMSE is decreased to 0.115, which is improved 11.03% and 16.20% respectively.

(3) The corn residue high covered area is identified by developing the U-Net convolution neural network, which is done on field level by harmazing the GF-1B/D and GF-2 PMS high spatial reslolution images.

The spatial resolutions of GF-1 B/D and GF-2 PMS images are high, which make potentional for depicting the detailed texture, spatial structure of the area with high corn residue coverage level. Deep learning can capture the inherent laws and representation levels of image sampling, which makes it huge advantages and great potential for identifying the corn residue cover on field. Therefore, this paper is exploring the method of identifying the area with high corn residue cover. Considering the image coverage and detail expression of remote sensing for imaging, the high spatial GF-2 PMS image are used to train the deep learning network for mining the high spatial resolution of image and detailed texture, pattern, and context of corn residue covered area. After that the trained network will be transferred to learn the land objects charaterritics of GF-1B/D PMS which is with high spatial resolution and wide image coverage at the same time. So, the high corn residue cover areas in the whole Lishu County can be identified only use one or two GF-1B/D PMS remote sensing images.

In the process of encoding and decoding, U-Net convolutional neural network can extract more high-level features of the larger receptive field by down-sampling, and capture more details of land objexts in image by up-sampling. Therefore, the U-Net convolutional neural network can capture the spatial details and spatial dimention on multi-scale. So, the U-Net convolutional neural network is used to capture the detailed textural, structural, and spectral features of GF-2 PMS and GF-1 B/D high spatial resolution images for identifying the area with high corn residue cover. The identified results indicate that the area with high corn residue cover can be identified accurately using U-Net convolutional neural network based on GF-2 PMS and GF-1B/D PMS images. The user accuracy and mapping accuracy are 94.18% and 93.12%, respectively. And the overall classification accuracy is 89.29%, and Kappa coefficient is 0.7853. Consequently, the U-Net convolutional neural network method can be used to identify the high corn residue covered area accurately based on GF-2 PMS and GF-1B/D PMS high spatial resolution images. This exploration is important for monitor the crop residue coverage and the amount of crop residue leaft in field.

(4) Proposing the method of estimating the coverage all year around in black soil by Chinese GaoFen series satellite images, and the estimation accuracy is validated in Lishu County, Jilin Province.

Based on the above exploration of estimating the fractional vegetation cover, crop residue cover and identifying the crop residue coverage area, the estimations of the coverage all year around of corn planted area in the whole Lishu County are done systematically. Lishu County is the model for protecting the black soil. And the Chinese GF1-WFV, GF-1 B/D PMS, GF-2 PMS and GF-6 WFV series satellite images are used for the monitor of the coverage all year around for black soil protection. The corn growing season in Lishu County is ranging from late May to late September. The vegetation cover estimation results indicate that the green vegetation coverage was lower on June 12 (< 0.15), which is higher than 0.75 in a long period from June 28 to September 6. The corn vegetation cover estimation results indicate the utilization of black soil in corn growing season. The corn residue cover estimation results indicate that the black soil is covered by corn residue from the middle and early October to the early May of the next year with the coverage of more than 0.3. The high corn residue cover identified result by GF-1B/D PMS and GF-2 PMS images show that there are about 135 825 hectares of residue high coverage area in Lishu County. All these above estimation results show that the black soil can be covered effectively all year around in Lishu County by applying the conservation tillage.

The cover monitoring results validation in Lishu County show that the Chinese GF series satellite images can be used to monitor the cover all year around in black soil accurately and timely. This exporation can be used to mine the potential of GF series satellite images in black soil protection, which can be generalized to other area.