The timely and accurate extraction of greenhouse instance (GI) is of significant practical importance to estimate the vegetable cultivation areas and yield prediction. Deep learnings (driven by knowledge learned from large-scale sample data) can be expected to adaptively explore the features of image data, compared with the traditional image analysis, such as unsupervised, supervised, and object-oriented classification. The end-to-end accurate extraction of GI information can also ensure the model generalization for the less manual intervention. However, there are still two challenges to identify the GI using deep learning. One is that the multiple GI can be mistakenly assumed as a continuous distribution in the dense areas of GI, leading to the segmentation errors. Another is that the degradation of performance could occur, when transferring the GI model to the large-scale spatial context. In this study, the region-boundary instance extraction was proposed using convolutional neural networks (CNNs) and morphological post-processing. At first, a region-boundary multi-class model was constructed to generate the boundary auxiliary labels of greenhouse. The network loss function was modified to enhance the boundary information recognition, and then to facilitate the removal of greenhouse boundaries from the recognition, thereby achieving the extraction of GI. Subsequently, the high-precision GI data was obtained using morphological operations, such as the instance object dilation and the minimum bounding rectangles. The high-resolution three-band remote sensing data was collected from Shouguang, Shandong Province, in order to train the base model. Three transfer modes (namely, pure transfer, scale-adaptive, and model fine-tuning mode) were then explored in five transfer research areas, including Xinjiang, Liaoning, Yunnan, Hubei, and Zhejiang Province. The accuracy of GI extraction was evaluated using common semantic segmentation performance metrics, unit intersection over union (UIoU), and instance recall rate (IRR). The research results indicate that the UNet was better suited to construct as the "Region-Boundary" multi-class model, compared with the semantic segmentation networks, such as PSPNet, DeeplabV3+, and HRNet. The higher semantic accuracy was achieved in the UNet with the UIoU and IRR of 2.43 and 2.91 percentage points higher, respectively, compared with the overall suboptimal HRNet. Furthermore, two morphological post-processing operations (instance dilation and the minimum bounding rectangle recognition) were simultaneously introduced to increase the UIoU and IRR by 10.53 and 1.44 percentage points, respectively. The scale adaptation mode was then adopted to adjust the input image resolution. The UIoU and IRR were improved from 3.02 to 22.71 percentage points, and from 2.40 to 30.33 percentage points, respectively, in all test datasets of migration areas, compared with the simple migration mode. The fine-tuning of the base model was utilized to adjust the input image resolution in the model fine-tuning mode. The UIoU and IRR were improved ranging from 37.76 to 50.67 percentage points, and from 46.04 to 76.87 percentage points. The higher accuracy was achieved in the GI recognition, with the UIoU and IRR of 13.64 and 14.18 percentage points higher than the conventional approaches, respectively. Simultaneously, the model transfer was applied to select the different migration modes, according to the scene differences between the predicted and training regions. The automated mapping of GI can be expected to efficiently and accurately extract the GI information over the large-scale areas. The finding can provide the information support to the intelligent construction of agricultural facilities.

Accurately and timely acquiring the planted area of crops can greatly contribute to grain yield estimation, market stability and food security. Remote sensing has been widely applied in agriculture, with the advancement of satellite remote sensing and mapping. However, the crop mapping is generally constrained by the insufficient number of representative training samples. In this study, a multi-stage sample purification was developed to integrate the temporal, spatial, spectral, and phenological information. The historical and spatial distribution of crops and current-season remote sensing images were utilized to automatically generate high-quality training samples. The crop was automatically identified. The multi-year crop distribution of the dataset was selected to extract the stable cultivation regions of target crops. Subsequently, morphological processing was conducted to minimize the plot edge effects. Area filtering was then used to avoid fragmentation for the large-scale planting areas of target crops. The random points were also generated for the sample points. Initially, spectral filtering was applied to the sample points to enhance their spectral purity. Then, the potential erroneous sample points were filtered using phenological features. Finally, the probability of random forest was used for the final selection, in order to obtain the ultimate data of sample points. The spatiotemporal reconstruction was conducted to filter the remote sensing data in a time series using the Google Earth Engine (GEE) cloud platform and Sentinel-2 data. The linear interpolation was employed with the Whittaker smoother. Subsequently, Random Forest was used to achieve the automated identification of early-season, one-season late, and double-cropping late rice in Wenzhou and Linhai City, Zhejiang Province, China. The feature bands included the reflectance data in three spectra (green, red, and near-infrared bands) and six vegetation indices, namely NDVI, LSWI, EVI, GCVI, NDTI, and NDSVI. The results indicate that sufficient training samples were generated using historical thematic data and current-season remote sensing images, thus achieving a sample accuracy as high as 98.5%. Experimental results indicate that the better performance of sample purification was achieved with the noticeably higher accuracy, compared with the rest. There was some impact of sample point quantity and image features on classification. Recognition accuracy was improved with an increase in the number of sample points. A plateau was reached when the number exceeded 3 000. Features in the irrigation period shared the most significant impact on rice identification. Random Forest model was utilized to obtain the accuracy of early-season rice over 96%, with a Kappa coefficient exceeding 0.93, and an F1 score exceeding 0.95. The planting areas of early, single-season late, and double-season late rice in Wuling City and Linhai City were taken as the ground truth values, in order to compare with the identified rice area. The high correlation was achieved with a coefficient of determination (R²) of 0.96. There was a high level of consistency between the identified area and the ground truth data. The high spatial resolution imagery showed that the rice identification exhibited high spatial detail and accuracy, with the accurate delineation of field boundaries. The improved model demonstrated high recognition accuracy, particularly in the early identification of rice. Additionally, the generation of an error map was simulated to evaluate the robustness of historical classification data with errors. This finding can be expected to identify other crops, such as corn, soybeans, and wheat. The efficiency of remote sensing crop areas was improved to reduce the impact of human factors. This finding can provide regional-level crop identification using sample generation, indicating the broad potential applications in large-scale automated crop mapping.