STRATEGIC OPTIMIZATION OF CONVOLUTIONAL NEURAL NETWORKS FOR HYPERSPECTRAL LAND COVER CLASSIFICATION

Abstract. Hyperspectral data recorded by future earth observation satellites will have up to hundreds of narrow bands that cover a wide range of the electromagnetic spectrum. The spatial resolution (around 30 meters) of such data, however, can impede the integration of the spatial domain for a classification due to spectrally mixed pixels and blurred edges in the data. Hence, the ability of performing a meaningful classification only relying on spectral information is important. In this study, a model for the spectral classification of hyperspectral data is derived by strategically optimizing a convolutional neural network (1D-CNN). The model is pre-trained and optimized on imagery of different nuts, beans, peas and dried fruits recorded with the Cubert ButterflEye X2 sensor. Subsequently, airborne hyperspectral datasets (Greding, Indian Pines and Pavia University) are used to evaluate the CNN's capability of transfer learning. For that, the datasets are classified with the pre-trained weights and, for comparison, with the same model architecture but trained from scratch with random weights. The results show substantial differences in classification accuracies (from 71.8% to 99.8% overall accuracy) throughout the used datasets, mainly caused by variations in the number of training samples, the spectral separability of the classes as well as the existence of mixed pixels for one dataset. For the dataset that is classified least accurately, the greatest improvement with pre-training is achieved (difference of 3.3% in overall accuracy compared to the non-pre-trained model). For the dataset that is classified with the highest accuracy, no significant transfer learning was observed.

Download Full-text

Tomato pest classification using deep convolutional neural network with transfer learning, fine tuning and scratch learning

Intelligent Decision Technologies ◽

10.3233/idt-200192 ◽

2021 ◽

pp. 1-10

Author(s):

Gayatri Pattnaik ◽

Vimal K. Shrivastava ◽

K. Parvathi

Keyword(s):

Neural Network ◽

Convolutional Neural Network ◽

Transfer Learning ◽

Data Augmentation ◽

State Of The Art ◽

Deep Convolutional Neural Network ◽

Fine Tuning ◽

Tomato Plants ◽

Random Weights

Pests are major threat to economic growth of a country. Application of pesticide is the easiest way to control the pest infection. However, excessive utilization of pesticide is hazardous to environment. The recent advances in deep learning have paved the way for early detection and improved classification of pest in tomato plants which will benefit the farmers. This paper presents a comprehensive analysis of 11 state-of-the-art deep convolutional neural network (CNN) models with three configurations: transfers learning, fine-tuning and scratch learning. The training in transfer learning and fine tuning initiates from pre-trained weights whereas random weights are used in case of scratch learning. In addition, the concept of data augmentation has been explored to improve the performance. Our dataset consists of 859 tomato pest images from 10 categories. The results demonstrate that the highest classification accuracy of 94.87% has been achieved in the transfer learning approach by DenseNet201 model with data augmentation.

Download Full-text

Recent Applications of Multispectral Imaging in Seed Phenotyping and Quality Monitoring—An Overview

Sensors ◽

10.3390/s19051090 ◽

2019 ◽

Vol 19 (5) ◽

pp. 1090 ◽

Cited By ~ 26

Author(s):

Gamal ElMasry ◽

Nasser Mandour ◽

Salim Al-Rejaie ◽

Etienne Belin ◽

David Rousseau

Keyword(s):

Quality Assessment ◽

Multispectral Imaging ◽

Seed Quality ◽

Hyperspectral Data ◽

Imaging Systems ◽

Quality Monitoring ◽

Electromagnetic Spectrum ◽

Safety Control ◽

Practical Applications ◽

Wide Range

As a synergistic integration between spectroscopy and imaging technologies, spectral imaging modalities have been emerged to tackle quality evaluation dilemmas by proposing different designs with effective and practical applications in food and agriculture. With the advantage of acquiring spatio-spectral data across a wide range of the electromagnetic spectrum, the state-of-the-art multispectral imaging in tandem with different multivariate chemometric analysis scenarios has been successfully implemented not only for food quality and safety control purposes, but also in dealing with critical research challenges in seed science and technology. This paper will shed some light on the fundamental configuration of the systems and give a birds-eye view of all recent approaches in the acquisition, processing and reproduction of multispectral images for various applications in seed quality assessment and seed phenotyping issues. This review article continues from where earlier review papers stopped but it only focused on fully-operated multispectral imaging systems for quality assessment of different sorts of seeds. Thence, the review comprehensively highlights research attempts devoted to real implementations of only fully-operated multispectral imaging systems and does not consider those ones that just utilized some key wavelengths extracted from hyperspectral data analyses without building independent multispectral imaging systems. This makes this article the first attempt in briefing all published papers in multispectral imaging applications in seed phenotyping and quality monitoring by providing some examples and research results in characterizing physicochemical quality traits, predicting physiological parameters, detection of defect, pest infestation and seed health.

Download Full-text

Rotation-Based Support Vector Machine Ensemble in Classification of Hyperspectral Data With Limited Training Samples

IEEE Transactions on Geoscience and Remote Sensing ◽

10.1109/tgrs.2015.2481938 ◽

2016 ◽

Vol 54 (3) ◽

pp. 1519-1531 ◽

Cited By ~ 50

Author(s):

Junshi Xia ◽

Jocelyn Chanussot ◽

Peijun Du ◽

Xiyan He

Keyword(s):

Support Vector Machine ◽

Hyperspectral Data ◽

Support Vector ◽

Training Samples ◽

Limited Training Samples

Download Full-text

A Efficient Solution for Classification of Crops using Hyper Spectral Satellite Images

International Journal of Innovative Technology and Exploring Engineering - Special Issue ◽

10.35940/ijitee.b7311.129219 ◽

2019 ◽

Vol 9 (2) ◽

pp. 5204-5211

Keyword(s):

Hyperspectral Image ◽

Satellite Image ◽

Hyperspectral Data ◽

Classification Model ◽

Research Activity ◽

Research Issues ◽

Computational Overhead ◽

Wide Range ◽

Crop Classification

Crop identification (CI) utilizing hyperspectral pictures/images (HSI) collected from satellite is one of the effective research area considering various agriculture related applications. Wide range of research activity is carried out and modelled in the area of crop recognition (CR) for building efficient model. Correlation filter (CF) is considered to be one of an effective method and are been applied by existing methodologies for identifying similar signal features. Nonetheless, very limited is work is carried out using CF for crop classification using hyperspectral data. Further, effective method is required that bring good tradeoffs between memory and computational overhead. The crop classification model can be improved by combining machine learning (ML) technique with CF. HSI is composed of hundreds of channels with large data dimension that gives entire information of imaging. Thus, using classification model is very useful for real-time application uses. However, the accuracy of classification task is affected as HSI is composed of high number of redundant and correlated feature sets. Along with, induce computational overhead with less benefits using redundant features. Thus, effective band selection, texture analysis, and classification method is required for accurately classifying multiple crops. This paper analyses various existing techniques for identification and classification of crops using satellite imagery detection method. Then, identify the research issues, challenges, and problems of existing model for building efficient techniques for identification and classification of crops using satellite image. Experiment are conducted on standard hyperspectral data. The result attained shows proposed model attain superior classification accuracy when compared with existing hyperspectral image classification model.

Download Full-text

Hyperspectral Image Classification Based on Parameter-Optimized 3D-CNNs Combined with Transfer Learning and Virtual Samples

Remote Sensing ◽

10.3390/rs10091425 ◽

2018 ◽

Vol 10 (9) ◽

pp. 1425 ◽

Cited By ~ 7

Author(s):

Xuefeng Liu ◽

Qiaoqiao Sun ◽

Yue Meng ◽

Min Fu ◽

Salah Bourennane

Keyword(s):

Transfer Learning ◽

Network Performance ◽

Hyperspectral Image ◽

Three Dimensional ◽

Feature Space ◽

Training Samples ◽

Virtual Samples ◽

3D Cnn ◽

Target Data

Recent research has shown that spatial-spectral information can help to improve the classification of hyperspectral images (HSIs). Therefore, three-dimensional convolutional neural networks (3D-CNNs) have been applied to HSI classification. However, a lack of HSI training samples restricts the performance of 3D-CNNs. To solve this problem and improve the classification, an improved method based on 3D-CNNs combined with parameter optimization, transfer learning, and virtual samples is proposed in this paper. Firstly, to optimize the network performance, the parameters of the 3D-CNN of the HSI to be classified (target data) are adjusted according to the single variable principle. Secondly, in order to relieve the problem caused by insufficient samples, the weights in the bottom layers of the parameter-optimized 3D-CNN of the target data can be transferred from another well trained 3D-CNN by a HSI (source data) with enough samples and the same feature space as the target data. Then, some virtual samples can be generated from the original samples of the target data to further alleviate the lack of HSI training samples. Finally, the parameter-optimized 3D-CNN with transfer learning can be trained by the training samples consisting of the virtual and the original samples. Experimental results on real-world hyperspectral satellite images have shown that the proposed method has great potential prospects in HSI classification.

Download Full-text

A l1-minimization Based Approach for Hyperspectral Data Classiﬁcation

Key Engineering Materials ◽

10.4028/www.scientific.net/kem.500.675 ◽

2012 ◽

Vol 500 ◽

pp. 675-681 ◽

Cited By ~ 4

Author(s):

Sami Ul Haq Qazi ◽

Li Xin Shi ◽

Lin Mi Tao ◽

Shi Qiang Yang

Keyword(s):

Sparse Representation ◽

Classification Problem ◽

Hyperspectral Data ◽

Model Learning ◽

Parameters Selection ◽

Learning Procedure ◽

Training Samples ◽

Supervised Methods ◽

First Time

Most of classiﬁcation methods require model learning procedure or optimal parameters selection using large number of training samples. In this paper, we propose a novel classiﬁcation approach using l1-minimization based sparse representation which does not need any learning procedure or parameters selection. The proposed approach is based on l1minimization because l0-minimization is generally NP-hard and is not a convex optimization problem. Sparse based solutions have been proposed in other areas like signal processing but to the best of our knowledge, this is the ﬁrst time that sparse based classiﬁcation is being proposed for the classiﬁcation of hyperspectral data. l1-minimization based sparse representation is calculated for each of the test samples using few training samples directly. In remote sensing, usually it is extremely difﬁcult and expensive to identify and label the samples due to which we often lack sufﬁcient training samples for classiﬁcation. We tested the proposed approach for difﬁcult classiﬁcation problem i.e high dimensional spaces and few training samples. We also analyzed the cases in which the sparse based classiﬁcation may and may not work. Our extensive experiments on real hyperspectral dataset (AVIRIS 1992 Indiana’s Indian Pines image), prove that the proposed technique offers more classiﬁcation accuracy and more efﬁcient than state-of-the-art SVM and semi-supervised methods as it does not need any model selection.

Download Full-text

Multilabel multiclass classification of OCT images augmented with age, gender and visual acuity data

10.1101/316349 ◽

2018 ◽

Cited By ~ 1

Author(s):

Parmita Mehta ◽

Aaron Lee ◽

Cecilia Lee ◽

Magdalena Balazinska ◽

Ariel Rokem

Keyword(s):

Neural Network ◽

Visual Acuity ◽

Macular Degeneration ◽

Transfer Learning ◽

Age Related Macular Degeneration ◽

Multilabel Classification ◽

Age Related ◽

Wide Range ◽

Retinal Pathologies

AbstractOptical Coherence Tomography (OCT) imaging of the retina is in widespread clinical use to diagnose a wide range of retinal pathologies and several previous studies have used deep learning to create systems that can accurately classify retinal OCT as indicative of one of these pathologies. However, patients often exhibit multiple pathologies concurrently. Here, we designed a novel neural network algorithm that performs multiclass and multilabel classification of retinal images from OCT images in four common retinal pathologies: epiretinal membrane, diabetic macular edema, dry age-related macular degeneration and neovascular age-related macular degeneration. Furthermore, clinicians often also use additional information about the patient for diagnosis. Second contribution of this study is improvement of multiclass, multilabel classification augmented with information about the patient: age, visual acuity and gender. We compared two training strategies: a network pre-trained with ImageNet was used for transfer learning, or the network was trained from randomly initialized weights. Transfer learning does not perform better in this case, because many of the low-level filters are tuned to colors, and the OCT images are monochromatic. Finally, we provide a transparent and interpretable diagnosis by highlighting the regions recognized by the neural network.

Download Full-text