Efficient Convolutional Neural Networks for Multi-Spectral Image Classification

Senecal, Jacob; Sheppard, John W.; Shaw, Joseph A.

doi:10.1109/ijcnn.2019.8851840

Cited by 19 publications

(16 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the mathematical computations of the Random Forest can be obtuse, we have tried in this paper to provide an intuitive explanation of its theoretical underpinnings, which we believe can be communicated more easily than those of, say, a Convolutional Neural Network (CNN) (LeCun et al, 2015). Additionally, most pre-trained CNNs have been built on RGB images (Senecal et al, 2019). In the future, researchers may yet develop CNNs that can analyze more than three bands and that would not require the computational resources typically demanded by deep CNNs.…”

Section: Discussionmentioning

confidence: 99%

Tracking Cloud Forests With Cloud Technology and Random Forests

Pascual¹,

Pascual²

2021

Front. Environ. Sci.

View full text Add to dashboard Cite

Hotspots of endemic biodiversity, tropical cloud forests teem with ecosystem services such as drinking water, food, building materials, and carbon sequestration. Unfortunately, already threatened by climate change, the cloud forests in our study area are being further endangered during the Covid pandemic. These forests in northern Ecuador are being razed by city dwellers building country homes to escape the Covid virus, as well as by illegal miners desperate for money. Between August 2019 and July 2021, our study area of 52 square kilometers lost 1.17% of its tree cover. We base this estimate on simulations from the predictive model we built using Artificial Intelligence, satellite images, and cloud technology. When simulating tree cover, this model achieved an accuracy between 96 and 100 percent. To train the model, we developed a visual and interactive application to rapidly annotate satellite image pixels with land use and land cover classes. We codified our algorithms in an R package—loRax—that researchers, environmental organizations, and governmental agencies can readily deploy to monitor forest loss all over the world.

show abstract

Section: Discussionmentioning

confidence: 99%

Tracking Cloud Forests With Cloud Technology and Random Forests

Pascual¹,

Pascual²

2021

Front. Environ. Sci.

View full text Add to dashboard Cite

show abstract

“…Previous results using SpectrumNet on multi-spectral images when compared to competing methods such as ResNet-50 39 and DenseNet-161 40 show the relative superiority of our approach, both in terms of accuracy and computational complexity. 30 100.0 ± 0, 00 100 ± 0.00…”

Section: Age Analysismentioning

confidence: 99%

“…For this step, we employed a convolutional neural network architecture that we call "Spectrum-Net." The details of this network are described in Senecal, 29 with a specific description of its application to related multispectral classification problems in Senecal et al 30 The basic architecture of SpectrumNet defines a set of "spectral" modules that correspond to squeeze layers and expand layers. The squeeze layers define the bulk of the network, except for a convolutional layer placed at the beginning and end of the network.…”

Section: Age Analysismentioning

confidence: 99%

Hyperspectral imaging and machine learning for monitoring produce ripeness

Logan¹,

Scherrer²,

Senecal³

et al. 2020

Sensing for Agriculture and Food Quality and Safety XII

Self Cite

View full text Add to dashboard Cite

Hyperspectral imaging is a powerful remote sensing tool capable of capturing rich spectral and spatial information. Although the origins of hyperspectral imaging are in terrestrial remote sensing, new applications are emerging rapidly. Owing to its non-destructive nature, hyperspectral imaging has become a useful tool for monitoring produce ripeness. This paper describes the process that uses a visible near-infrared (VNIR) hyperspectral imager from Resonon, Inc., coupled with machine learning algorithms to assess the ripeness of various pieces of produce. The images were converted to reflectance across a spectral range of 387.12 nm to 1023.5 nm, with a spectral resolution of 2.12 nm. A convolutional neural network was used to perform age classification for potatoes, bananas, and green peppers. Additionally, a genetic algorithm was used to determine the wavelengths carrying the most useful information for age classification. Experiments were run using RGB images, full spectrum hyperspectral images, and the genetic algorithm feature selection method. Results showed that the genetic algorithm-based feature selection method outperforms RGB images for all tested produce, outperforms hyperspectral imagery for bananas, and matches hyperspectral imagery performance for green peppers. This feature selection method is being used to develop a low-cost multi-spectral imager for use in monitoring produce in grocery stores.

show abstract

“…The results of our proposed model are also compared to those of three supervised methods: ResNet50, 48 SpectrumNet, 59 and HybridSN. 21 ResNet50 was selected for this comparison because it constitutes a deep convolutional network that has achieved excellent generalization performance on many image classification tasks.…”

Section: #Correctly Classified Observations #Total Observations ;mentioning

confidence: 99%

Reduced-cost hyperspectral convolutional neural networks

Morales

Sheppard

Scherrer

et al. 2020

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

Hyperspectral imaging provides a useful tool for extracting complex information when visual spectral bands are not enough to solve certain tasks. However, processing hyperspectral images (HSIs) is usually computationally expensive due to the great amount of both spatial and spectral data they incorporate. We present a low-cost convolutional neural network designed for HSI classification. Its architecture consists of two parts: a series of densely connected three-dimensional (3-D) convolutions used as a feature extractor, and a series of twodimensional (2-D) separable convolutions used as a spatial encoder. We show that this design involves fewer trainable parameters compared to other approaches, yet without detriment to its performance. What is more, we achieve comparable state-of-the-art results testing our architecture on four public remote sensing datasets: Indian Pines, Pavia University, Salinas, and EuroSAT; and a dataset of Kochia leaves [Bassia scoparia] with three different levels of herbicide resistance. The source code and datasets are available online. (Hyper3DNet codebase: https://github.com/GiorgioMorales/hyper3dnet.) © 2020 Society of Photo-Optical Instrumentation Engineers (SPIE)

show abstract

Efficient Convolutional Neural Networks for Multi-Spectral Image Classification

Cited by 19 publications

References 10 publications

Tracking Cloud Forests With Cloud Technology and Random Forests

Tracking Cloud Forests With Cloud Technology and Random Forests

Hyperspectral imaging and machine learning for monitoring produce ripeness

Reduced-cost hyperspectral convolutional neural networks

Contact Info

Product

Resources

About