Plants and weeds discrimination model

• Project Status:: On going..
• Projec

Section

• Overview
• Working Flowchart
• Model break down

Overview

Goal:

• Design a robust and simple to integrate plants and weeds machine learning model.
• Proposes plants/weeds discriminant model that are able to process in real time.

Why?

• Machine learning vision is wide area of research. Automated Crops and weed control will save cost and reduce environment impact.

Working Flowchart

Model break down

Dataset

• This image dataset has 15336 segments, being 3249 of soil, 7376 of soybean, 3520 grass and 1191 of broadleaf weeds.
• We want the model to uniformly learn all classes therefore,
  • training set: 4000 images(1000 each classes)
  • validation set: 200 images(50 each classes)
  • test: 200(50 each classes)

Difference Learning model on Dataset

• Determine plant coordinate in input images/video
  • Plant blob coordinate
• Preprocessing and get Familiar with Dataset
  • 10% of Dataset is testset
• Problem:
  • limited dataset. with roughly 10,000 more images each model below will achieve 7.5 - 15% more

Model	Best Accuracy	Note	Status
Logistic Regression	59.5%
HSV space - 2 Layer net	34.5%	Bad Overall
2 Layer Net	76.5%(Not stable)
Fully Connected Net	78.5%(More stable)

References

• Link to dataset

Keyword

Kaggle | Neural net | Open CV | Pytorch | Jupyter notebook | Python3