Abstract
With the development of remote sensing technology, the semantic segmentation and recognition of various things in the ocean have become more and more frequent. Due to the wide variety of marine things and the large differences in morphology, it has brought greater difficulties to the recognition of marine remote sensing images. In order to obtain better segmentation results of ocean remote sensing images, this paper proposes an cross attention mechanism(Horizontal and Vertical) of exponential operation combined with multi-scale convolution algorithm. Among them, the cross attention mechanism and expanded distribution weight coefficient mentioned in this paper are first proposed. First, Input the marine remote sensing image features into an cross attention mechanism algorithm of exponential operation to obtain feature weight coefficients and joint weight coefficients in multiple directions; Then, the features with weight coefficients are input into the multi-access convolutional layer and the multi-scale dilated convolutional layer respectively for deep feature mining; Then the above steps are repeated twice, and finally the semantic segmentation of marine remote sensing images is achieved by fusing multiple deep-level features afterwards. Experiments were conducted on three public marine remote sensing data sets, and the results proved the effectiveness of our proposed cross attention mechanism of extended operation algorithm. The F values of the MAMC model on Beach, Island and Sea ice data sets have reached 99.4%, 91.25%, 87.08% respectively. Compared with other models, the effect is significantly improved, and proved the powerful performance of the algorithm in the semantic segmentation of marine remote sensing images.
Highlights
Since the development of remote sensing technology, this technology has been applied to various fields, such as image identification [1], [2], [36], scene classification [3]–[5], and semantic segmentation [6]–[8]
To solve the above problems, we propose a cross attention mechanism with an exponential operation combined with a multiscale convolution (MAMC) algorithm that plays an active role in eliminating redundant features and deep feature mining
The model MAMC proposed in this paper jointly mines shallow features through multi-scale convolution and multi-scale dilated convolution to obtain more deep features, so the effect is the best
Summary
Since the development of remote sensing technology, this technology has been applied to various fields, such as image identification [1], [2], [36], scene classification [3]–[5], and semantic segmentation [6]–[8]. Research on marine remote sensing has become increasingly important [9]–[14]. Due to the variety of marine things and the large differences in their morphological attributes, the remote sensing images after imaging are complicated and diverse. In order to better semantically segment marine remote sensing images, we need to fully mine the deep features of remote sensing images and eliminate a large number of redundant features. Researchers have proposed some new methods to solve redundant features and choose some deep features.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
