Previously:

• 230507-2308 230507-1623 Plants paper notes
• 230529-1453 Plant datasets taxonomy prep

Comparative Analysis of Deep Learning Techniques for “in-the-wild” Plant Classification

Abstract

Introduction

• For almost two decades, machine learning has been successfully(?) applied to plant classification and related tasks, from the use of hand-crafted features with SVM and Random Forest for tasks like leaf¹ and flower² identification, to the Computer Vision and Deep Learning techniques predominant in recent (WHICH? EXAMPLES?) years that achieved impressive results on a wide variety of tasks.
  • For example, in agriculture, Deep Learning (DL) is employed for crop/plant/fruit recognition/counting, disease detection, yield estimation, classification - not just by species/taxon but also by seed presence, freshness, phenology (e.g. maturity), etc.: <@mamatAdvancedTechnologyAgriculture2022 Advanced Technology in Agriculture Industry by Implementing Image Annotation Technique and Deep Learning Approach (2022) z/d>
• But one of the most interesting synergies happened, and is happening, at the intersection of ML, citizen science, and biodiversity.
• Biodiversity loss
  • TODO: is bad
    • Biodiversity can be defined as “a measure of variation at the genetic (genetic variability), species (species diversity), and ecosystem (ecosystem diversity) level” ³
  • In the context of

Motivation

Biodiversity

• Biodiversity monitoring helps biodiversity
  • Helps tracking the situation
  • Helps prioritization of resources

Citizen science

• Is cool
• Really nice for reasons such as
  • getting people interested in science
  • wisdom of the crowd
  • if the crowd doesn’t have much wisdom, many pair of eyes can still be good/useful!
• Really nice if you don’t really know what to track, or want to analyze data post-factum
  • As opposed to clear experiments where you take expensive scientists to track specific plant types

Main contribution

The main contributions of this paper are:

• A review of existing literature on the topic of using ML for plant species classification
• Establish a baseline

Related work

Supervised Pretraining

Our results with ViT nicely match with page 9 of <@picekPlantRecognitionAI2022 Plant recognition by AI (2022) z/d>