Usefulness of interpretability methods to explain deep learning based plant stress phenotyping

Thumbnail Image
Nagasubramanian, Koushik
Singh, Arti
Sarkar, Soumik
Major Professor
Committee Member
Journal Title
Journal ISSN
Volume Title
Singh, Asheesh
Research Projects
Organizational Units
Organizational Unit
Mechanical Engineering
The Department of Mechanical Engineering at Iowa State University is where innovation thrives and the impossible is made possible. This is where your passion for problem-solving and hands-on learning can make a real difference in our world. Whether you’re helping improve the environment, creating safer automobiles, or advancing medical technologies, and athletic performance, the Department of Mechanical Engineering gives you the tools and talent to blaze your own trail to an amazing career.
Organizational Unit
Electrical and Computer Engineering

The Department of Electrical and Computer Engineering (ECpE) contains two focuses. The focus on Electrical Engineering teaches students in the fields of control systems, electromagnetics and non-destructive evaluation, microelectronics, electric power & energy systems, and the like. The Computer Engineering focus teaches in the fields of software systems, embedded systems, networking, information security, computer architecture, etc.

The Department of Electrical Engineering was formed in 1909 from the division of the Department of Physics and Electrical Engineering. In 1985 its name changed to Department of Electrical Engineering and Computer Engineering. In 1995 it became the Department of Electrical and Computer Engineering.

Dates of Existence

Historical Names

  • Department of Electrical Engineering (1909-1985)
  • Department of Electrical Engineering and Computer Engineering (1985-1995)

Related Units

Organizational Unit

The Department of Agronomy seeks to teach the study of the farm-field, its crops, and its science and management. It originally consisted of three sub-departments to do this: Soils, Farm-Crops, and Agricultural Engineering (which became its own department in 1907). Today, the department teaches crop sciences and breeding, soil sciences, meteorology, agroecology, and biotechnology.

The Department of Agronomy was formed in 1902. From 1917 to 1935 it was known as the Department of Farm Crops and Soils.

Dates of Existence

Historical Names

  • Department of Farm Crops and Soils (1917–1935)

Related Units

Journal Issue
Is Version Of

Deep learning techniques have been successfully deployed for automating plant stress identification and quantification. In recent years, there is a growing push towards training models that are interpretable -i.e. that justify their classification decisions by visually highlighting image features that were crucial for classification decisions. The expectation is that trained network models utilize image features that mimic visual cues used by plant pathologists. In this work, we compare some of the most popular interpretability methods: Saliency Maps, SmoothGrad, Guided Backpropogation, Deep Taylor Decomposition, Integrated Gradients, Layer-wise Relevance Propagation and Gradient times Input, for interpreting the deep learning model. We train a DenseNet-121 network for the classification of eight different soybean stresses (biotic and abiotic). Using a dataset consisting of 16,573 RGB images of healthy and stressed soybean leaflets captured under controlled conditions, we obtained an overall classification accuracy of 95.05 \%. For a diverse subset of the test data, we compared the important features with those identified by a human expert. We observed that most interpretability methods identify the infected regions of the leaf as important features for some -- but not all -- of the correctly classified images. For some images, the output of the interpretability methods indicated that spurious feature correlations may have been used to correctly classify them. Although the output explanation maps of these interpretability methods may be different from each other for a given image, we advocate the use of these interpretability methods as `hypothesis generation' mechanisms that can drive scientific insight.


This is a pre-print of the article Nagasubramanian, Koushik, Asheesh K. Singh, Arti Singh, Soumik Sarkar, and Baskar Ganapathysubramanian. "Usefulness of interpretability methods to explain deep learning based plant stress phenotyping." arXiv preprint arXiv:2007.05729 (2020). Posted with permission.

Wed Jan 01 00:00:00 UTC 2020