Enhancing ensemble diversity based on multiscale dilated convolution in image classification

You, Gui-Rong; Shiue, Yeou-Ren; Su, Chao‐Ton; Huang, Qing-Lan

doi:10.1016/j.ins.2022.05.064

Cited by 19 publications

(7 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We also experimented with different convolution methods and ensemble strategies to augment segmentation performance. Dilated convolution has been proposed to increase the receptive field of a network, helping to extract more global and higher-level semantic features (You et al 2022 ). We did not observe a significant improvement in performance between dilated and conventional convolution methods (DSC p = 0.947).…”

Section: Discussionmentioning

confidence: 99%

Head and neck tumor segmentation convolutional neural network robust to missing PET/CT modalities using channel dropout

et al. 2023

View full text Add to dashboard Cite

ObjectiveRadiation therapy for Head and Neck (H&N) cancer relies on accurate segmentation of the primary tumor. A robust, accurate, and automated gross tumor volume segmentation method is warranted for H&N cancer therapeutic management. The purpose of this study is to develop a novel deep learning segmentation model for H&N cancer based on independent and combined CT and FDG-PET modalities.ApproachIn this study, we developed a robust deep learning-based model leveraging information from both CT and PET. We implemented a 3D U-Net architecture with 5 levels of encoding and decoding, computing model loss through deep supervision. We used a channel dropout technique to emulate different combinations of input modalities. This technique prevents potential performance issues when only one modality is available, increasing model robustness. We implemented ensemble modeling by combining two types of convolutions with differing receptive fields, conventional and dilated, to improve capture of both fine details and global information.Main ResultsOur proposed methods yielded promising results, with a Dice Similarity Coefficient (DSC) of 0.802 when deployed on combined CT and PET, DSC of 0.610 when deployed on CT, and DSC of 0.750 when deployed on PET.SignificanceApplication of a channel dropout method allowed for a single model to achieve high performance when deployed on either single modality images (CT or PET) or combined modality images (CT and PET). Furthermore, ensemble modeling showed comparable or improved performance by combining advantages of conventional and dilated convolution, while decreasing associated generalization errors. The presented segmentation techniques are clinically relevant to applications where images from a certain modality might not always be available.

show abstract

Section: Discussionmentioning

confidence: 99%

Head and neck tumor segmentation convolutional neural network robust to missing PET/CT modalities using channel dropout

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Ensemble learning [58] is an effective way to improve task performance and is widely used in deep networks [52,53,59,60]. Our approach differs from existing ensemble methods [61,62], which focus on the accuracy of ID data, while our approach focuses on detecting OOD data. In addition, our model is an ensemble of different smoothing priors, which outperforms general deep integration methods.…”

Section: Ensemble Learningmentioning

confidence: 99%

Out‐of‐distribution detection based on multi‐classifiers

Jiang

2023

Cognitive Comp and Systems

View full text Add to dashboard Cite

Existing out‐of‐distribution detection models rely on the prediction of a single classifier and are sensitive to classifier bias, making it difficult to discriminate similar feature out‐of‐distribution data. This article proposed a multi‐classifier‐based model and two strategies to enhance the performance of the model. The model first trains several different base classifiers and obtains the predictions of the test data on each base classifier, then uses cross‐entropy to calculate the dispersion between these predictions, and finally uses the dispersion as a metric to identify the out‐of‐distribution data. A large scatter implies inconsistency in the predictions of the base classifier, and the greater the probability of belonging to the out‐of‐distribution data. The first strategy is applied in the training process of the model to increase the difference between base classifiers by using various scales of Label smoothing regularisation. The second strategy is applied to the inference process of the model by changing the mean and variance of the activations in the neural network to perturb the inference results of the test data. These two strategies can effectively amplify the discrepancy in the dispersion of the in‐distribution and out‐of‐distribution data. The experimental results show that the method in this article can effectively improve the performance of the model in the detection of different types of out‐of‐distribution data, improve the robustness of deep neural networks (DNN) in the face of unknown classes, and promote the application of DNN in systems and engineering with high security requirements.

show abstract

“…Time-frequency images have clear boundaries, with distinct signal and non-signal regions, strong continuity between adjacent pixel values, and representation information that is reflected in a strong spatial structure connection method between pixel points. In contrast, natural images exhibit spatial correlation between adjacent pixels, where this correlation is localized to only surrounding pixels and not the entire image [26].…”

Section: A the Comparison And Analysis Of Tfi And Natural Imagesmentioning

confidence: 99%

“…Given natural images contain a significant amount of subtle information crucial for classification, smaller convolution kernels can locally perceive images and extract their detailed features. Using smaller convolution kernels also results in lower computational complexity, enabling faster operation [26,29], therefore, they are better suited for natural image recognition. However, it is evident that smaller receptive fields of convolution kernels are inadequate to extract comprehensive spatial structural features [30], which is necessary for TFI.…”

Section: B Multi-scale Dilated Residual Networkmentioning

confidence: 99%

“…Aiming at the aforementioned objectives, we propose a novel design idea of the Dilated Convolution Neural Networks (DCNN) inspired by the idea of dilated residual network [25], multi-scale fusion [26], and denoise [27,28]. On the one hand, in this research, we employed dilated convolution to increase the receptive field and capture more comprehensive structural features.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Dilated CNN Design Approach for Extracting Multi-Scale Features in Radar Emitter Classification

Guo,

Wu,

Guo

et al. 2023

IEEE Access

View full text Add to dashboard Cite

Radar emitter classification plays an increasingly significant role in the electronic reconnaissance system. Due to many convolutional neural network (CNN)-based approaches suffer from insufficient spatial receptive fields and inadequate feature representation, the classification accuracy is poor in low signal-to-noise ratio (SNR) conditions. Therefore, in this paper, we stress the importance of multiscale dilated convolutions for target feature extraction, and propose two novel CNN architecture design approaches called multi-scale dilated residual network (MDRN). By combining multi-scale dilated convolutions with residual architecture, MDRN not only has a larger receptive field, but also can learn more diverse features, thereby improving the ability to process time-frequency images (TFI) under highnoise energy conditions. Moreover, compared with the original residual model, MDRN does not increase any parameter complexity or floating-point operations per second (FLOPS). Experiments on the TFI classification task show that the proposed MDRN has superior performance over state-of-the-art CNN models.INDEX TERMS intra-pulse modulation classification; dilated convolution; feature maps; multi-scale feature fusion; time-frequency image

show abstract

Enhancing ensemble diversity based on multiscale dilated convolution in image classification

Cited by 19 publications

References 23 publications

Head and neck tumor segmentation convolutional neural network robust to missing PET/CT modalities using channel dropout

Head and neck tumor segmentation convolutional neural network robust to missing PET/CT modalities using channel dropout

Out‐of‐distribution detection based on multi‐classifiers

Dilated CNN Design Approach for Extracting Multi-Scale Features in Radar Emitter Classification

Contact Info

Product

Resources

About