HybridCNN based hyperspectral image classification using multiscale spatiospectral features

Mohan, Alkha; Venkatesan, M.

doi:10.1016/j.infrared.2020.103326

Cited by 69 publications

(25 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Recently, deep neural networks have become popular, following their successful application for classic image classification [ 33 , 34 ] and object detection [ 35 ]. Examples of architectures employed for HSI classification include one [ 10 ], two [ 36 ] or three-dimensional [ 22 ] convolutional models for more efficient use of spatial and spectral information [ 37 ].…”

Section: Introductionmentioning

confidence: 99%

Blood Stain Classification with Hyperspectral Imaging and Deep Neural Networks

Książek

Romaszewski

Głomb

et al. 2020

Sensors

View full text Add to dashboard Cite

In recent years, growing interest in deep learning neural networks has raised a question on how they can be used for effective processing of high-dimensional datasets produced by hyperspectral imaging (HSI). HSI, traditionally viewed as being within the scope of remote sensing, is used in non-invasive substance classification. One of the areas of potential application is forensic science, where substance classification on the scenes is important. An example problem from that area—blood stain classification—is a case study for the evaluation of methods that process hyperspectral data. To investigate the deep learning classification performance for this problem we have performed experiments on a dataset which has not been previously tested using this kind of model. This dataset consists of several images with blood and blood-like substances like ketchup, tomato concentrate, artificial blood, etc. To test both the classic approach to hyperspectral classification and a more realistic application-oriented scenario, we have prepared two different sets of experiments. In the first one, Hyperspectral Transductive Classification (HTC), both a training and a test set come from the same image. In the second one, Hyperspectral Inductive Classification (HIC), a test set is derived from a different image, which is more challenging for classifiers but more useful from the point of view of forensic investigators. We conducted the study using several architectures like 1D, 2D and 3D convolutional neural networks (CNN), a recurrent neural network (RNN) and a multilayer perceptron (MLP). The performance of the models was compared with baseline results of Support Vector Machine (SVM). We have also presented a model evaluation method based on t-SNE and confusion matrix analysis that allows us to detect and eliminate some cases of model undertraining. Our results show that in the transductive case, all models, including the MLP and the SVM, have comparative performance, with no clear advantage of deep learning models. The Overall Accuracy range across all models is 98–100% for the easier image set, and 74–94% for the more difficult one. However, in a more challenging inductive case, selected deep learning architectures offer a significant advantage; their best Overall Accuracy is in the range of 57–71%, improving the baseline set by the non-deep models by up to 9 percentage points. We have presented a detailed analysis of results and a discussion, including a summary of conclusions for each tested architecture. An analysis of per-class errors shows that the score for each class is highly model-dependent. Considering this and the fact that the best performing models come from two different architecture families (3D CNN and RNN), our results suggest that tailoring the deep neural network architecture to hyperspectral data is still an open problem.

show abstract

Section: Introductionmentioning

confidence: 99%

Blood Stain Classification with Hyperspectral Imaging and Deep Neural Networks

Książek

Romaszewski

Głomb

et al. 2020

Sensors

View full text Add to dashboard Cite

show abstract

“…Bu evrişim işlemi, giriş görüntüsünden uzamsal özellikleri çıkarmaya yardımcı olmaktadır. Her nöronun 2B evrişim çıktısı Denklem (1)'deki gibi formüle edilmektedir (Mohan and Venkatesan, 2020).…”

Section: öNerilen Yöntemunclassified

“…3B evrişim, 3B spektral görüntülerden uzamsal-spektral özelliklerin çıkarılmasına yardımcı olmaktadır. 3B ESA yönteminden çıkarılan özellik Denklem (2)'deki gibi formüle edilmektedir (Mohan and Venkatesan, 2020).…”

Section: öNerilen Yöntemunclassified

“…Derin öğrenme tekniklerindeki ilerleme, HUAG sınıflandırmasını yeni bir aşamaya taşımaktadır. Derin öğrenme yönteminin eğitim aşaması, özellikleri otomatik olarak çıkarmakta ve bunları sınıflandırma aşamasında kullanmaktadır (Mohan and Venkatesan, 2020). Tipik bir derin öğrenme modeli olarak, yığınlı otokodlayıcı (Zhang et al, 2016) ve derin inanç ağı (Chen et al, 2015) hem uzamsal hem de spektral bilgiyi çıkarabilmekte ve daha sonra bunları HUAG sınıflandırması için birleştirebilmektedir.…”

Section: Introductionunclassified

See 1 more Smart Citation

Hibrid 3B-2B ESA Mimarisi Kullanılarak Hiperspektral Uzaktan Algılama Görüntülerinin Sınıflandırılması

FIRAT¹,

Uçan²,

Hanbay³

2021

Computer Science

View full text Add to dashboard Cite

Özetçe-Hiperspektral uzaktan algılama görüntüleri (HUAG) iki uzamsal ve bir spektral boyuta sahip görüntü küpleridir. Evrişimsel sinir ağı (ESA), HUAG sınıflandırmada uzamsal-spektral özellik bilgilerinin çıkarılmasını sağlayan en etkili derin öğrenme yöntemlerinden biridir. Geleneksel ESA tabanlı yöntemler genellikle özellik çıkarımı için 2B ESA'yı kullanmaktadır. 2B ESA yalnızca uzamsal boyutttaki özellikleri yakaladığından dolayı tek başına spektral boyutlardan iyi ayırt edici özellik haritaları çıkaramaz. 3B ESA HUAG'deki uzamsal-spektral özellikleri eş zamanlı olarak çıkarabilmektedir. Ancak 3B ESA hesaplama açısından karmaşıktır. Bu çalışmada 3B-2B ESA birleşiminden oluşan hibrid bir yöntem önerilmiştir. 3B ESA ortak uzamsal-spektral özellikleri çıkarırken, 3B ESA'dan sonra kullanılan 2B ESA ile daha fazla uzamsal özellikler öğrenilir. Ayrıca önerilen hibrid yöntem hesaplama karmaşıklığını da azaltmaktadır. Önerilen yöntemin sınıflandırma performansını arttırmak için Mish aktivasyon fonksiyonu kullanılmaktadır. Yaygın olarak kullanılan iki veriseti ile gerçekleştirilen uygulamalar sonucunda önerilen yöntemin daha iyi sınıflandırma performansına sahip olduğu görülmektedir.

show abstract

“…Recently, a CNN model has been used successfully in hyperspectral image analysis. Mohan and Venkatesan 24 developed a hybrid CNN model for hyperspectral image classification and this achieved good results on the ‘Indian pine’, ‘Pavia university’ and ‘Salinas’ datasets. In a study with crops, Yang et al 25 .…”

Section: Introductionmentioning

confidence: 99%

Identification of the variety of maize seeds based on hyperspectral images coupled with convolutional neural networks and subregional voting

et al. 2021

View full text Add to dashboard Cite

BACKGROUND Maize is one of the most important food crops in the world. Many different varieties of maize seeds are similar in size and appearance, so distinguishing the varieties of maize seed is a significant research topic. This study used hyperspectral image processing coupled with convolutional neural network (CNN) and a subregional voting method to recognize different varieties of maize seed. RESULTS First, visible and near‐infrared (NIR‐visible) hyperspectral images were obtained. Savitzky–Golay (SG) smoothing and first derivative (FD) were used to pretreat the raw spectra and highlight the spectral differences of samples of different varieties. Second, the region of interest (ROI) of each sample was divided into several subregions according to the shape and the number of pixels. Then, a method was proposed for reshaping the images of pixel spectra for the CNN and the training model was established. Finally, using subregional voting, one prediction result was generated from the prediction results of several original subregions in one sample. The results showed that, for six varieties of normal maize seeds, the tests identified embryoid and non‐embryoid forms with 93.33% and 95.56% accuracy, respectively. For six varieties of sweet maize seeds, the test accuracy in embryoid and non‐embryoid forms was 97.78% and 98.15%, respectively. CONCLUSION The maize seed was identified accurately. The present study demonstrated that the CNN model for spectral image coupled with subregional voting represents a new approach for the identification of varieties of maize seed. © 2021 Society of Chemical Industry

show abstract

HybridCNN based hyperspectral image classification using multiscale spatiospectral features

Cited by 69 publications

References 20 publications

Blood Stain Classification with Hyperspectral Imaging and Deep Neural Networks

Blood Stain Classification with Hyperspectral Imaging and Deep Neural Networks

Hibrid 3B-2B ESA Mimarisi Kullanılarak Hiperspektral Uzaktan Algılama Görüntülerinin Sınıflandırılması

Identification of the variety of maize seeds based on hyperspectral images coupled with convolutional neural networks and subregional voting

Contact Info

Product

Resources

About