RS-DARTS: A Convolutional Neural Architecture Search for Remote Sensing Image Scene Classification

Abstract

Due to the superiority of convolutional neural networks, many deep learning methods have been used in image classification. The enormous difference between natural images and remote sensing images makes it difficult to directly utilize or modify existing CNN models for remote sensing scene classification tasks. In this article, a new paradigm is proposed that can automatically design a suitable CNN architecture for scene classification. A more efficient search framework, RS-DARTS, is adopted to find the optimal network architecture. This framework has two phases. In the search phase, some new strategies are presented, making the calculation process smoother, and better distinguishing the optimal and other operations. In addition, we added noise to suppress skip connections in order to close the gap between trained and validation processing and ensure classification accuracy. Moreover, a small part of the neural network is sampled to reduce the redundancy in exploring the network space and speed up the search processing. In the evaluation phase, the optimal cell architecture is stacked to construct the final network. Extensive experiments demonstrated the validity of the search strategy and the impressive classification performance of RS-DARTS on four public benchmark datasets. The proposed method showed more effectiveness than the manually designed CNN model and other methods of neural architecture search. Especially, in terms of search cost, RS-DARTS consumed less time than other NAS methods.

AUTO-HAR: An adaptive human activity recognition framework using an automated CNN architecture design

“…The Neural Architecture Search (NAS) technique, which automates the design of deep learning networks, has been shown to perform equally well or better than manually designed architectures in a variety of computer vision tasks compared to conventional deep learning techniques that rely heavily on expert or domain knowledge [19] , [59] . Therefore, several different search strategies have been proposed, including evolutionary deep learning (EDL) [12] , random search methods [20] , gradient-based methods [21] , and Bayesian optimization methods [22] Another factor to consider in architectural search is the scope of the search space. Two search spaces are available: the entire architecture for a given set of inputs and outputs [20] , [23] , or the inner structure (operations and connections) of a fixed macro architecture [24] , [25] , which is not optimized during the architecture search and functions as an exoskeleton.…”

Innovative Hyperspectral Image Classification Approach Using Optimized CNN and ELM

“…Cai et al [57] presented a novel cross-attention mechanism and graph convolution integration algorithm. Zhang et al [58] presented a convolutional neural architecture for remote sensing image scene classification. Hilal et al [59] presented a new deep transfer learning-based fusion model for remote-sensing image classification.…”

Multimodal Semantic Consistency-Based Fusion Architecture Search for Land Cover Classification

“…Despite promising performance, the major issues of existing MDL methods with intermediately fusion mainly contain two aspects. On the one hand, duo to the close-range images in ImageNet dataset and the RS images are very different in terms of imaging angles and imaging methods, various handcrafted backbones are used into the MDL methods that may not be the optimal solution in the RS image interpretation problems [23], Many researchers [24], [25] have demonstrated that pre-trained handcraft-based RS interpretation methods have been difficult to further improve due to the limitations of manually designed network architecture, which proposed a series of convolutional neural architecture search frameworks to automatically search the network structure for the RS images, and obtain better Fig. 1.…”

“…Recently, NAS strategies have been investigated into the intelligent interpretation of the RS images [23], [24], [25], [38], [39], which aims to apply the characteristics of the RS data to guide the computer to automatically learn network structure oriented to the RS domain, and then solving the limitations of manually designed network architecture oriented to the natural images. Liu et al, [24] proposed a continuous particle swarm optimization-based deep learning architecture search model, which automatically design cellbased CNN architectures and obtained better performance over latest hyper-spectral images (HIS) classification methods based on handcrafted CNN.…”

“…Liu et al, [24] proposed a continuous particle swarm optimization-based deep learning architecture search model, which automatically design cellbased CNN architectures and obtained better performance over latest hyper-spectral images (HIS) classification methods based on handcrafted CNN. Duo to the enormous difference between natural images and the RS images, it is inappropriate that directly migrating natural images-oriented existing CNN models for the RS images processing, Zhang et al, [25] proposed a more efficient search framework, RS-DARTS, to find the scene classification-oriented network architecture. Furthermore, to balance computational complexity and performance error of searched network, a framework for scene classification network architecture search based on multi-objective neural evolution (SceneNet) was proposed [38].…”