TPENAS: A Two-Phase Evolutionary Neural Architecture Search for Remote Sensing Image Classification

Ao, Lei; Feng, Kaiyuan; Sheng, Kai; Zhao, Hongyu; He, Xin; Chen, Zigang

doi:10.3390/rs15082212

Cited by 8 publications

(6 citation statements)

References 75 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hence, the results of this research could be discussed with respect to studies carried out in forests of different growing stages. An improvement of the OA of species classification between 1-4.3 pp was documented in mature forests by hybridizing the CNN and k-nearest neighbors [54], combining Res-Net and U-net structures [55], using their proposed 3D-1D-CNN method [56], and using a novel two-phase CNN [57]. Hence, the magnitude of improving the OA of classifying seedlings in this study was in line with the aforementioned literature of mature forests.…”

Section: Discussionmentioning

confidence: 99%

A New Approach for Feeding Multispectral Imagery into Convolutional Neural Networks Improved Classification of Seedlings

Imangholiloo,

Luoma,

Holopainen

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Tree species information is important for forest management, especially in seedling stands. To mitigate the spectral admixture of understory reflectance with small and lesser foliaged seedling canopies, we proposed an image pre-processing step based on the canopy threshold (Cth) applied on drone-based multispectral images prior to feeding classifiers. This study focused on (1) improving the classification of seedlings by applying the introduced technique; (2) comparing the classification accuracies of the convolutional neural network (CNN) and random forest (RF) methods; and (3) improving classification accuracy by fusing vegetation indices to multispectral data. A classification of 5417 field-located seedlings from 75 sample plots showed that applying the Cth technique improved the overall accuracy (OA) of species classification from 75.7% to 78.5% on the Cth-affected subset of the test dataset in CNN method (1). The OA was more accurate in CNN (79.9%) compared to RF (68.3%) (2). Moreover, fusing vegetation indices with multispectral data improved the OA from 75.1% to 79.3% in CNN (3). Further analysis revealed that shorter seedlings and tensors with a higher proportion of Cth-affected pixels have negative impacts on the OA in seedling forests. Based on the obtained results, the proposed method could be used to improve species classification of single-tree detected seedlings in operational forest inventory.

show abstract

Section: Discussionmentioning

confidence: 99%

A New Approach for Feeding Multispectral Imagery into Convolutional Neural Networks Improved Classification of Seedlings

Imangholiloo,

Luoma,

Holopainen

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…(2022a) proposed a new CNN consisting of a group of so-called self-compensating convolutional modules. Ao et al. (2023), Broni-Bediako et al.…”

Section: Related Workmentioning

confidence: 99%

“…E.g., Shi et al (2022a) proposed a new CNN consisting of a group of so-called self-compensating convolutional modules. Ao et al (2023), Broni-Bediako et al (2022 and Shen et al (2022a, b) proposed three CNN architectures obtained by neural architecture search. Speaking of advantages, these four methods both have fewer parameters compared to the off-theshelf CNNs developed on ImageNet-1K.…”

Section: Related Workmentioning

confidence: 99%

MBC-Net: long-range enhanced feature fusion for classifying remote sensing images

Song

2023

IJICC

View full text Add to dashboard Cite

PurposeClassification of remote sensing images (RSI) is a challenging task in computer vision. Recently, researchers have proposed a variety of creative methods for automatic recognition of RSI, and feature fusion is a research hotspot for its great potential to boost performance. However, RSI has a unique imaging condition and cluttered scenes with complicated backgrounds. This larger difference from nature images has made the previous feature fusion methods present insignificant performance improvements.Design/methodology/approachThis work proposed a two-convolutional neural network (CNN) fusion method named main and branch CNN fusion network (MBC-Net) as an improved solution for classifying RSI. In detail, the MBC-Net employs an EfficientNet-B3 as its main CNN stream and an EfficientNet-B0 as a branch, named MC-B3 and BC-B0, respectively. In particular, MBC-Net includes a long-range derivation (LRD) module, which is specially designed to learn the dependence of different features. Meanwhile, MBC-Net also uses some unique ideas to tackle the problems coming from the two-CNN fusion and the inherent nature of RSI.FindingsExtensive experiments on three RSI sets prove that MBC-Net outperforms the other 38 state-of-the-art (STOA) methods published from 2020 to 2023, with a noticeable increase in overall accuracy (OA) values. MBC-Net not only presents a 0.7% increased OA value on the most confusing NWPU set but also has 62% fewer parameters compared to the leading approach that ranks first in the literature.Originality/valueMBC-Net is a more effective and efficient feature fusion approach compared to other STOA methods in the literature. Given the visualizations of grad class activation mapping (Grad-CAM), it reveals that MBC-Net can learn the long-range dependence of features that a single CNN cannot. Based on the tendency stochastic neighbor embedding (t-SNE) results, it demonstrates that the feature representation of MBC-Net is more effective than other methods. In addition, the ablation tests indicate that MBC-Net is effective and efficient for fusing features from two CNNs.

show abstract

“…Therefore, applying NAS to RSI data sets could yield more efficient deep models that leverage general RSI features. However, existing NAS studies (Ao et al , 2023; Broni-Bediako et al , 2022; Shen et al , 2022a) have only validated their models on RSI data sets significantly smaller than ImageNet-1K. Consequently, these NAS-based models have demonstrated smaller volumes without achieving competitive accuracy.…”

Section: Related Workmentioning

confidence: 99%

“…Alternatively, custom-built CNN architectures developed through neural architecture search (NAS) on RSI data sets may offer improved solutions. For instance, Ao et al (2023), Broni-Bediako et al (2022) and Shen et al (2022a) proposed their NAS-based CNNs and demonstrated much smaller volumes than preexisting models. However, these NAS-based methods still achieved significantly lower accuracy because the NAS was only performed on RSI data sets with several tens of thousands of data points.…”

Section: Introductionmentioning

confidence: 99%

Efficient knowledge distillation for remote sensing image classification: a CNN-based approach

Song,

Wei,

Yong

2023

IJWIS

View full text Add to dashboard Cite

Purpose The paper aims to tackle the classification of Remote Sensing Images (RSIs), which presents a significant challenge for computer algorithms due to the inherent characteristics of clustered ground objects and noisy backgrounds. Recent research typically leverages larger volume models to achieve advanced performance. However, the operating environments of remote sensing commonly cannot provide unconstrained computational and storage resources. It requires lightweight algorithms with exceptional generalization capabilities. Design/methodology/approach This study introduces an efficient knowledge distillation (KD) method to build a lightweight yet precise convolutional neural network (CNN) classifier. This method also aims to substantially decrease the training time expenses commonly linked with traditional KD techniques. This approach entails extensive alterations to both the model training framework and the distillation process, each tailored to the unique characteristics of RSIs. In particular, this study establishes a robust ensemble teacher by independently training two CNN models using a customized, efficient training algorithm. Following this, this study modifies a KD loss function to mitigate the suppression of non-target category predictions, which are essential for capturing the inter- and intra-similarity of RSIs. Findings This study validated the student model, termed KD-enhanced network (KDE-Net), obtained through the KD process on three benchmark RSI data sets. The KDE-Net surpasses 42 other state-of-the-art methods in the literature published from 2020 to 2023. Compared to the top-ranked method’s performance on the challenging NWPU45 data set, KDE-Net demonstrated a noticeable 0.4% increase in overall accuracy with a significant 88% reduction in parameters. Meanwhile, this study’s reformed KD framework significantly enhances the knowledge transfer speed by at least three times. Originality/value This study illustrates that the logit-based KD technique can effectively develop lightweight CNN classifiers for RSI classification without substantial sacrifices in computation and storage costs. Compared to neural architecture search or other methods aiming to provide lightweight solutions, this study’s KDE-Net, based on the inherent characteristics of RSIs, is currently more efficient in constructing accurate yet lightweight classifiers for RSI classification.

show abstract

TPENAS: A Two-Phase Evolutionary Neural Architecture Search for Remote Sensing Image Classification

Cited by 8 publications

References 75 publications

A New Approach for Feeding Multispectral Imagery into Convolutional Neural Networks Improved Classification of Seedlings

A New Approach for Feeding Multispectral Imagery into Convolutional Neural Networks Improved Classification of Seedlings

MBC-Net: long-range enhanced feature fusion for classifying remote sensing images

Efficient knowledge distillation for remote sensing image classification: a CNN-based approach

Contact Info

Product

Resources

About