Unsupervised Parameterization for Optimal Segmentation of Agricultural Parcels from Satellite Images in Different Agricultural Landscapes

Abstract

Image segmentation is a cost-effective way to obtain information about the sizes and structural composition of agricultural parcels in an area. To accurately obtain such information, the parameters of the segmentation algorithm ought to be optimized using supervised or unsupervised methods. The difficulty in obtaining reference data makes unsupervised methods indispensable. In this study, we evaluated an existing unsupervised evaluation metric that minimizes a global score (GS), which is computed by summing up the intra-segment uniformity and inter-segment dissimilarity within a segmentation output. We modified this metric and proposed a new metric that uses absolute difference to compute the GS. We compared this proposed metric with the existing metric in two optimization approaches based on the Multiresolution Segmentation (MRS) algorithm to optimally delineate agricultural parcels from Sentinel-2 images in Lower Saxony, Germany. The first approach searches for optimal scale while keeping shape and compactness constant, while the second approach uses Bayesian optimization to optimize the three main parameters of the MRS algorithm. Based on a reference data of agricultural parcels, the optimal segmentation result of each optimization approach was evaluated by calculating the quality rate, over-segmentation, and under-segmentation. For both approaches, our proposed metric outperformed the existing metric in different agricultural landscapes. The proposed metric identified optimal segmentations that were less under-segmented compared to the existing metric. A comparison of the optimal segmentation results obtained in this study to existing benchmark results generated via supervised optimization showed that the unsupervised Bayesian optimization approach based on our proposed metric can potentially be used as an alternative to supervised optimization, particularly in geographic regions where reference data is unavailable or an automated evaluation system is sought.

Evaluation of Global-Scale and Local-Scale Optimized Segmentation Algorithms in GEOBIA With SAM on Land Use and Land Cover

“…In computer vision tasks, the intersection over union (IoU) is a commonly employed evaluation index. Furthermore, the F measure can also be utilized for assessing the segmentation results of deep learning methods [45]. Given that SAM only segments some geographical objects in the image, it is unreasonable to apply evaluation indicators other than F and IoU to assess its segmentation results.…”

“…The second category of segmentation methods is supervised segmentation, which requires the use of labeled training data to guide the algorithm for learning, such as Graph Cut [41], Conditional Random Field [42], and deep learning methods [43], [44], [45]. In recent years, the persistent exploration and refinement of deep learning have been motivated by the remarkable achievements of convolutional neural networks and the explosion of novel architectures.…”

Evaluation of Sentinel-1 and Sentinel-2 Feature Sets for Delineating Agricultural Fields in Heterogeneous Landscapes

“…This layer was used to create a mask to remove non-agricultural areas from the MMC images before segmenting the agricultural fields. This approach has also been used in other studies [31], [36], [37].…”

“…Building on the experiences of the Landsat and SPOT missions, Sentinel-2 (S2) was designed within the framework of the European Copernicus program for land surface and agriculture monitoring [28] at a temporal resolution of 5 days and a spatial resolution of 10 m. As opposed to optical sensors, which are inhibited by clouds, Sentinel-1 (S1), which is also part of the Copernicus program enables the continuous monitoring of the earth's surface in all weather conditions at a temporal resolution of 6 days and a spatial resolution of 20 m. Various researchers have used S1 [29], [30], and predominantly S2 [31]- [40] for segmenting agricultural fields. In using the S1 or S2 images, most of those authors used existing segmentation algorithms (e.g., [29], [30], [32]- [34], [36]- [40]), some proposed new segmentation algorithms (e.g., [31], [35]), and others proposed new segmentation parameter optimization approaches (e.g., [36], [37]). One area that is yet to be comprehensively explored is the determination of the optimal feature set from S1 and S2 images for segmenting agricultural fields given that both sensors come with different bands and additional features like band indices can be calculated as well.…”