"No one is harder on a talented person than the person themselves" - Linda Wilkinson ; "Trust your guts and don't follow the herd" ; "Validate direction not destination" ;

January 04, 2019

Day #179 - Machine learning in plant breeding - Alencar Xavier's PhD defense - Already Startups are their in this Area :) - Good One Liked it

Key Summary
  • ML - Part of AI for pattern recognition for problem solving, Prediction, Classification and sometimes inference
  • Heritability using the animal model
ML in Plant Breeding
  • Intelligent Decision making
  • Maximize productivity, Minimize losses
  • Plant breeding is about resource allocation 


ML Use cases (Very Interesting)
  • Yields
  • Select Target environments (Plants behave similarly)
  • Genomic-based prediction
  • Soil Classification

Dataset
  • SoyNam
  • Agronomic traits

Feature Variables (So many variables we can collect from a plant :), Awesome and Amazing )
  • Yield Days
  • Days to Flowering
  • Days to maturity
  • Length of Reproductive period
  • Lodging Score (Scale 1 to 5)
  • Average Canopy Closure (From Pictures)
  • Rate of Canopy Closure
  • Leaflet Shape
  • Node and Pod Number in the main stem
  • Pods per node
  • Internal Length - Height / Number of Nodes
Analytics Steps
  • Check for Multicollinearity
  • Genomic relationship matrix
Unsupervised ML Use Case
  • Cluster among traits
  • Genetics / Interaction based
  • Pearson correlation
  • Phenotypic correlation
  • Conditional independence (Markov random fields)
  • Finding the connected traits
  • Genetic and Environmental Correlations map


Patterns Identified


PCA Approach
  • Reduction of Dimensionality
  • Eigenvectors of matrix
  • Similar story in PCA
Supervised ML (Prediction)
  • Adjust Phenotypes
  • Kriging
  • Genotyping Density (Posterior distribution)




Related Paper #1 - 
From paper - Soil Data Analysis Using Classification Techniques and Soil Attribute Prediction

Key Soil Traits

Regression Use case - Predicting - Phosphorous attribute on least number of attributes.

Related Paper #2 - From paper A real-time phenotyping framework using machine learning for plant stress severity rating in soybean

Use cases - Image to identify Plant Rating, the Classification problem

The Image Processing Key Tasks
  • From the images extract white balance and color calibration, Hue Values
  • Segmentation RGB, background removal
  • Noise and outlier removal


Related Paper #3 - DeepWheat: Estimating Phenotypic Traits from Crop Images with Deep Learning (DeepWheat_Estimating_Phenotypic_Traits_From_Images.pdf)
  • Phenotypic - observable characteristics - made manually in the field
  • Aerial images as a basis for biomass estimation
  • Above-ground biomass (AGB)
  • Emergence count and biomass estimation frameworks using CNN
  • Randomized minimal region swapping (RMRS) algorithm for Biomass Estimation
Key Steps
  • Extract all the segmented patches from the whole image 
  • Input each patch image to the counting module to get the individual emergence counts for each patch
  • Predicted counts for a single plot image
Good CNN Summarization - a simple stack of convolutional layers with an over-complete set of filters followed by nonlinearity and pooling serve well when the underlying concepts to be learned via abstract representation are linearly separable



  • Our goal is to find the more suitable deep-learning architecture for our task
  • Handcrafted feature methods are the Histogram of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), which are usually combined with classifiers such as Adaptive Boosting (AdaBoost)  or Support Vector Machines (SVM)

Patterns Detection
  • Infection status
  • Location of the symptom
  • Patterns of the leaf
  • Type of fungus
  • Color and Shape



My Thoughts - This post has provided fundamentals on areas worked upon on Agri Startups.
  • Data Collection and infrastructure is key
  • Feature Variables are from Soil traits, Plant Traits, Weather Data
  • The Analysis of collected data for different periods/years would provide qualitative information to predict / forecast / recommend
  • Quality Assessment can also be predicted based on the trends
More Reads
Connected Components Labeling
Orange & Brown
Google Citation - Alencar Xavier



Happy Mastering DL!!!

No comments: