-
Notifications
You must be signed in to change notification settings - Fork 0
/
docs.typ
51 lines (37 loc) · 2.49 KB
/
docs.typ
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
#import("typst/template_paper.typ"): *
#import "@preview/tablex:0.0.5": tablex, cellx, hlinex, vlinex
// #set math.equation(numbering: "(1)")
// #set figure.caption(position: top)
// #show raw.where(block: true): block.with(
// fill: luma(250),
// stroke: luma(100) + 1pt,
// inset: 6pt,
// radius: 4pt,
// )
// #show quote: set align(center)
#show: paper.with(
title: [Crop Pest and Pathogen Detection with Computer Vision],
authors: (
"Connacher Murphy",
),
date: datetime.today().display("[month repr:long] [day], [year]"),
)
= Introduction
We develop a crop pest and pathogen diagnostic model. We train a neural network on the Mensah et al. (2023) CCMT dataset.
Our diagnostic model powers the `CroPP` tool at #link("https://saxifrage.co")[saxifrage].
= Data
We use the Mensah et al. (2023) CCMT dataset. We assign the images into training and test sets with probabilities 80% and 20%, respectively.
The data include labelled images of corn, cassava, maize, and tomatoes. We use only the maize data at present.
// == Maize (create a table with class frequencies)
= Architecture
We begin with a pre-trained instance of `resnet18`. We then conduct further training for each crop. We adopt a cross entropy loss function.
= Cross-Validation Exercises
We consider the selection of hyperparameters through cross-validation. We provide summary statistics in `validation_summary.pdf`.
= Deployment Training
The live version of `CroPP` is trained with
- $30$ epochs,
- a $1 times 10^(-4)$ learning rate,
- and a batch size of $128$.
You can view a demonstration of the detection model on the #link("https://saxifrage.co")[saxifrage website].
// The leaf spot mislabels:
// /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm835_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm802_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm803_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm804_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm805_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm806_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm832_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm833_.jpg /Users/connormurphy/data/ccmt/Raw Data/CCMT Dataset/Maize/fall armyworm/fall armyworm834_.jpg