Deep learning: A new engine for ecological resource research
Deep learning is a big data-driven machine learning method that can automatically be extracting complex high-dimensional nonlinear features. Although deep learning has achieved better performance for big data mining than traditional statistical learning and machine learning algorithms, there are still huge challenges when processing ecological resource data, including multi-source/multi-meta heterogeneity, spatial-temporal coupling, geographic correlation, high dimensional complexity, and low signal-to-noise ratio. A recent study clarified the aforementioned frontier issues. The related research paper entitled "The Application of Deep Learning in the Field of Ecological Resources Research: Theory, Methods, and Challenges" has been published in "Science in China: Earth Science".
Deep learning has made significant progress in many fields with the accumulation of data, the improvement of computing power, and the progress of algorithms. This study focuses on the application of deep learning in the field of ecological resources. The main contents include:
1) An overview of the history, development, and basic structure of deep learning (Figure 1). The relationships between ecological resource big data research and deep learning structures represented by convolutional neural networks, recurrent neural networks, and graph neural networks were analyzed (Figure 2)
2) The main tasks of deep learning, common public data sets, and tools in ecological resources were summarized (Figure 2).
3) The application of deep learning in plant image classification, crop phenotype, and vegetation mapping was demonstrated. The application ability and potential of deep learning in structured and unstructured ecological data were analyzed.
4) The challenges and prospects of deep learning in the application of ecological resources were analyzed (Figure 3), including standardization and sharing of data, construction of crowdsourcing collection platform, interpretability of deep neural network, hybrid deep learning with domain knowledge, small sample learning, data fusion, and enrichment and intelligence of applications.
(Content and Image Courtesy: https://www.eurekalert.org/pub_releases/2020-05/scp-dla052120.php)