首页 > , Vol. , Issue () : -
本文旨在研究基于地块数据约束的深度学习模型的分类特征表示方法，以识别不同作物在不同时相上光谱差异从而对作物类型进行分类。通过Google Earth Engine平台获取作物生育期内全部Landsat-8影像，利用其质量评定波段完成研究区无云时相及区域上的地块统计，提取地块级别的各波段反射率均值按照时相顺序及波长进行排列，构建波谱、时相二维特征图作为该地块的抽象表示。通过构建相对最优的CNN结构完成对特征图的分类，从而完成对地块的分类。构建CNN模型并不需要手工特征和预定义功能的需求，可完成提取特征并遵循端到端原则进行分类。将该模型的分类结果与其他最为常用机器学习分类器进行了比较，获得了优于常用遥感分类算法的分类精度。结果表明地块数据的加入可以有效的缩减计算规模并提供了准确的分类边界。所提出得方法在地块特征表示及作物分类中具有突出的应用潜力，应视为基于地块的多时相影像分类任务的优选方法。
Crop planting area and its spatiotemporal distribution information are very important. It is of great significance for agricultural management, structural adjustment of planting industry, national food security, etc. Remote sensing can extract crop planting information quickly. However, in order to meet the actual production needs, a large number of ground survey information, expert knowledge and manual correction operation after classification are needed. Using remote sensing and existing geographic information data for intelligent information extraction is the future development trend. The purpose of this paper is to study the classification representation method of deep learning model based on the constraints of field parcel data. multi-source heterogeneous data are linked by constructing representation. The method can identify the spectral differences of different crops in different phases and classify the crop types. The research area of this paper is in the eighth division of Xinjiang production and Construction Corps. The platform of Google Earth engine is used to obtain all landsat-8 images during the crop growth period of the research area in 2019. Using image data and its quality assessment band on GEE to complete the statistical or estimated values of phases of the study area. Then, the average reflectivity of each band of the extracted block level is arranged according to the time-phase sequence and wavelength, and the plot representation of the spectrum and time-phase two-dimensional properties is constructed. The construction of land plot representation has completed the connection of geographic information data and remote sensing data, and made it possible to carry out the application of deep learning in crop remote sensing classification. The following convolution neural network model construction and training work are completed by using the completed plot representation. Through the step-by-step optimization process to search for the best combination of super parameters and various types of layers. The construction of the relatively optimal CNN model is completed, and the classification of the constructed feature map is carried out based on this model, so as to complete the classification of the research area in 2019. We have obtained the crop distribution map which is much higher than the resolution of remote sensing image. Secondly, the overall accuracy of CNN model reaches 93.04% Kappa coefficient reaches 91.09%. The results are good in nine kinds of crop classification. Moreover, after thousands of rounds of data learning, compared with other machine learning algorithms, this method has lower classification error fluctuation and stronger stability The results show that the research object is plot representation, which is the abstract expression of plot planting information. It can be used as the standard input of deep learning model. Through the construction of plot representation, crop classification can be indirectly identified by remote sensing. The proposed method has outstanding application potential inland feature representation and crop classification and should be regarded as the optimal method for multi-temporal image classification tasks based on field parcel data. This method can be used for reference in the application of deep learning in remote sensing.