Automatic Raft Labeling for Remote Sensing Images via Dual-Scale Homogeneous Convolutional Neural Network

Abstract

Raft-culture is a way of utilizing water for farming aquatic product. Automatic raft-culture monitoring by remote sensing technique is an important way to control the crop's growth and implement effective management. This paper presents an automatic pixel-wise raft labeling method based on fully convolutional network (FCN). As rafts are always tiny and neatly arranged in images, traditional FCN method fails to extract the clear boundary and other detailed information. Therefore, a homogeneous convolutional neural network (HCN) is designed, which only consists of convolutions and activations to retain all details. We further design a dual-scale structure (DS-HCN) to integrate higher-level contextual information for accomplishing sea-land segmentation and raft labeling at the same time in a uniform framework. A dataset with Gaofen-1 satellite images was collected to verify the effectiveness of our method. DS-HCN shows a satisfactory performance with a better interpretability and a more accurate labeling result.

Extracting Raft Aquaculture Areas from Remote Sensing Images via an Improved U-Net with a PSE Structure

“…In order to verify the effectiveness of our proposed raft aquaculture areas extraction method, we compare the UPS-Net on the selected dataset with several state-of-the-art methods, i.e., SVM [40], Mask R-CNN [22], FCN [19], U-Net [20], PSPNet [21], DeepLabv3+ [16] and DS-HCN [13]. The SVM method uses a radial basis kernel function [40] with two important parameters, C and γ, which were set to 1.8 and 0.34 respectively in the experiment.…”

“…The proposed UPS-Net has higher adaptability in raft aquaculture areas extraction because it can automatically fuse multiscale feature maps. DS-HCN is also an outstanding FCN-based method to extract raft aquaculture areas, which includes two fixed scales branches [13]. The small-scale branch is a homogeneous convolutional network with a receptive field of 33 × 33.…”

“…Hu et al modified statistical region-merging algorithm for superpixels segmentation and used fuzzy compactness and separation clustering algorithm to recognize raft aquaculture areas from SAR images [12]. However, the parameters of these methods have to be adjusted and analyzed manually [13]. Therefore, the accuracy of raft aquaculture areas extraction is largely dependent on lengthy parameter investigations and professional experience.…”

“…Since the boundaries in the boundary probability map of the RCF are often broken, it is necessary to repair the extracted vector boundary. Shi et al design a dual-scale homogeneous convolutional neural network (DS-HCN), which combines a small receptive field and a large receptive field to mine the association between multiple targets [13]. DS-HCN achieved a satisfactory performance on aquaculture areas exaction of Dalian, China.…”

Utilizing Multilevel Features for Cloud Detection on Satellite Imagery

“…We have moved into a Big Data Era [1,2], and an enormous amount of data are expected to be processed to accomplish different special tasks. With rapid remote sensing technology development springing up, optical satellite images are widely used for automatical applications, such as different applications of target detection [3][4][5][6][7][8] and scene classification [9]. However, clouds cover more than 50% of the surface of the earth [10][11][12], and consequently, clouds might be great challenges when automatically processing the images.…”