Extraction and Classification of the Supervised Coastal Objects Based on HSRIs and a Novel Lightweight Fully Connected Spatial Dropout Network

Abstract

For the protection and management of coastal ecosystems, it is crucial to monitor typical coastal objects and examine their characteristics of spatial and temporal variation. There are limitations to the conventional object-oriented and spectrum-based approaches to HSRIs interpretation. The majority of recently conducted studies on semantic segmentation based on DCNNs concentrate on improving the accuracy of single objects at local scales. The completeness, generalization, and edge accuracy of the extraction and classification of multiple objects with the complex background at regional scales still need to be improved. We created a benchmark dataset CSRSD for coastal supervision using HSRIs and GIS in this study to address the aforementioned problems. In the meantime, by combining the traditional U-Net and DeepLabV3+ feature fusion strategies, we propose a novel fully connected fusion pattern by switching to deepwise separable convolution from conventional convolution and introducing spatial dropout to create a brand new CBS module. The LFCSDN, a new lightweight fully connected spatial dropout network, has been suggested. The findings demonstrate that our constructed semantic segmentation dataset, which has produced reliable results on U-Net and DeepLabV3+, can be used as a benchmark for applications based on DCNNs for coastal scenes. While maintaining high accuracy, LFCSDN can significantly reduce the number of parameters. Our suggested CBS module can increase the model’s generalization by reducing overfitting. In order to analyze the spatiotemporal characteristics of target changes in the study area, tests on expansive remote sensing imagery were also conducted. The findings can be applied to ecological restoration, coastal area mapping, and integrated management. Additionally, it serves as a resource for studies on multiscale semantic segmentation in computer vision.

HR_BMNet: a boundary perception and multiresolution fusion network for semantic segmentation of remote sensing

“…In this section, we compared the proposed HR-BMNet with the classical state-of-the-art semantic segmentation methods. [1] 69.03 84.70 81.10 87.42 91.25 75.17 83.33 68.34 HRNet [11] 69.24 84.89 81.27 87.58 90.62 75.98 83.99 68.19 DeeplabV3+ [14] 68.01 84.71 80.21 87.04 90.76 76.02 83.89 63.33 SCAttNetV2 [15] 70.20 85.47 82.06 89.13 90.30 80.04 80.31 70.50 VPA [16] 68.77 84.80 83.94 87.12 90.65 76.24 83.94 66.52 FSAUNet [17] 69 FSAUNet incorporates feature self-attention during the encoding stage to effectively highlight the building region. As evident from the table, it per-forms exceptionally well in segmenting the building category.…”

“…This demonstrates our method's ability to effectively balance the segmentation results for different targets in urban scenes, ultimately yielding superior overall performance. [1] 87.36 98.17 93.97 93.29 94.34 85.52 99.01 LFCSDN [13] 87.13 98.24 90.56 93.12 92.59 85.99 -DeeplabV3+ [14] 87 [15] 88.11 98.38 93.49 93.77 94.42 86.65 99.12 VPA [16] 86.51 98.35 92.58 93.48 90.67 87.08 99.11 FSAUNet [17] 88.21 98.41 93.56 93.64 94.29 87.20 99.14 LightFGCNet [18] 88 To assess the generalization of our proposed method, we conducted a comparison with existing semantic segmentation methods using the CSRSD dataset. The table 3 highlights that FSAUNet achieves the best segmentation performance for the background category, with an F1 score of 99.14%, but only 0.02% higher than HR-BMNet.…”

Intelligent Detection of Marine Offshore Aquaculture with High-Resolution Optical Remote Sensing Images

“…Wang et al [18] combined multi-scale feature fusion and spatial rule induction to extract raft cultivation areas using high-spatial-resolution remote sensing imagery. It should be pointed out that ML requires considerable engineering and domain expertise to design a feature extractor that transforms raw data to features, such as spectral, textural, geometric relationship characteristics, so that the classifier can detect and classify different land covers [19,20]. Thus, for ML, defining features is necessary and multiple adjustments are required to achieve favorable results [21,22].…”

“…Some researchers have even proposed new deep learning methods for aquaculture area extraction. Chen et al [20] proposed a new lightweight fully connected spatial dropout network (LFCSDN), combining U-Net and DeepLab-v3+ feature fusion strategies to detect floating raft aquaculture areas. LFCSDN achieved an optimal F1 score of 94.04%, surpassing U-Net (91.64%) and DeepLab-v3+ (91.66%) [20].…”

“…Chen et al [20] proposed a new lightweight fully connected spatial dropout network (LFCSDN), combining U-Net and DeepLab-v3+ feature fusion strategies to detect floating raft aquaculture areas. LFCSDN achieved an optimal F1 score of 94.04%, surpassing U-Net (91.64%) and DeepLab-v3+ (91.66%) [20]. Deng et al [25] proposed a coastal aquaculture network based on GaoFen-2 satellite images to extract information about cage and raft aquaculture areas in mainland China, with overall accuracy of 94.64%.…”