An Effective Image Denoising Method for UAV Images via Improved Generative Adversarial Networks

Abstract

Unmanned aerial vehicles (UAVs) are an inexpensive platform for collecting remote sensing images, but UAV images suffer from a content loss problem caused by noise. In order to solve the noise problem of UAV images, we propose a new methods to denoise UAV images. This paper introduces a novel deep neural network method based on generative adversarial learning to trace the mapping relationship between noisy and clean images. In our approach, perceptual reconstruction loss is used to establish a loss equation that continuously optimizes a min-max game theoretic model to obtain better UAV image denoising results. The generated denoised images by the proposed method enjoy clearer ground objects edges and more detailed textures of ground objects. In addition to the traditional comparison method, denoised UAV images and corresponding original clean UAV images were employed to perform image matching based on local features. At the same time, the classification experiment on the denoised images was also conducted to compare the denoising results of UAV images with others. The proposed method had achieved better results in these comparison experiments.

Deriving Sea Subsurface Temperature Fields from Satellite Remote Sensing Data Using a Generative Adversarial Network Model

“…In deep generative networks, some data generation losses are optimized in a grid-based manner (grid-to-grid), so generated data typically lack high-frequency details; thus, the perceptual difference between the real data and the generated data is not understood (Kiasari et al, 2017;Wang et al, 2018). Our study area covers a large area of China's marginal seas, including the BHS, YS and ECS, with a high spatial resolution.…”

Spectrum of Advancements and Developments in Multidisciplinary Domains for Generative Adversarial Networks (GANs)

“…Wang et al [27] 2018 Noise filtering to avoid information loss during remote sending Benefit: image de-noising Qiuhong et al [28] 2019 High volume data redundancy by applying compression NLP-natural language processing (NLP) Li et al [29] 2018 Text regression model for association of text data and social outcome Advantage: data analysis with limited labelling Lin et al [30] 2017 Rank-Gan for data analysis and quality assessment using rank metric Qian et al [31] 2018 Event factuality identification using Ac-GAN by learning syntactic inform and address imbalance among factuality values Advantage: reduced reliance over annotated text Health Che et al [32] 2017 Hergan for synthetic health data generation with limited electronic health record (HER) Hwang et al [33] 2017 Disease prediction using AC-GAN and stacked auto-encoder Rezaei et al [34] 2018 Semantic segmentation and disease classification by selective weighted loss Advantage: address Data imbalance Fake audio, video and image generation Choi et al [35] 2018 Stargan for fake image generation by using deep CNN Achieved high classification accuracy Nataraj et al [36] 2019 Detection of fake images using co-occurrence matrices along with deep learning Achieved good generalization and very high classification accuracy Agriculture Suarez et al [37] 2017 Strength assessment of vegetation against normalized difference vegetation index (NDI) by applying Conditional GAN Barth et al [38] 2017 Cyclegan for gap reduction between synthetic and empirical image data set Advantage: ease of translation of color and textures Music Yang et al [39] 2017 Midinet-generation of musical notes by using CNN GAN Comparison of midinet was also made with Google's melodyrnn from scratch Advantage: combine existing melodies as well as generate melodies from multiple channels [40] 2018 Misegan-generates symbolic music i.e. piano-rolls of five tracks and four bars i.e.…”

“…layouts generation of relational graphic elements to wireframe images by modelling geometric relations of different types of two dimensional elements Advantage: introduction of wireframe rendering layer which produce a set of relational graphic controls Liu et al [62] 2018 Treegan for source code generation Advantage: syntax-aware sequence generation Fault prediction Gao et al [63] 2019 ASM1D-GAN a model to identify the faults by extracting features related to faults from real fault samples and create the similar one Advantage: integration of data creation and fault determination Zhou et al [64] 2019 Synthesize vibrational fault samples using a technique of global optimization Advantage: feature extraction of feature using limited number of samples and its effective representation using auto-encoder Filter the non-compliant synthetic samples which are not useful for reliable fault diagnosis Zheng et al [65] 2019 Gan-fp utilizes multiple GANs to create training samples and an inference network in parallel to predict failures for newly crafted samples Improved performance as well as significant socio-economic impact Text generation Subramanian et al [66] 2018 Ability to create sentence outlines using an adversarial model which learns the distribution of sentences in a hidden space persuaded by sentence encoder Advantage: produce real like samples with multinomial sampling Liang et al [67] 2017 Create useful distractors Advantage: achieves comparable performance to a frequently used word2vec-based method for the Wiki dataset Malware detection Dahl et al [68] 2013 Employ random projections to decrease the dimension of the original latent space Achieved improved classification results commerce and in capturing data for remote sensing images. Wang et al [27] the authors try to solve the problem of noise in UAVs. Noise causes loss of content during remote sensing of images by UAVs.…”

Synthesis of Prostate MR Images for Classification Using Capsule Network-Based GAN Model

“…The typical example is the Generative Adversarial Networks (GAN) introduced by Goodfellow et al [ 10 ]. Recently, many GAN-based models and their extensions have facilitated tackling the challenging image analysis problems such as segmentation [ 11 ], denoising [ 12 ], registration [ 13 ], detection [ 14 ] and classification [ 15 ]. As regards deep learning-based image classification, GAN has been widely for synthesizing training samples because its unique ability in mimicking data distributions indeed opens the possibility to bridge the gap between learning data and synthetic data [ 8 ].…”