Development of Machine Learning Models for Early Detection of Crop Diseases Based on Crop Imagery Data

This code base contains the code used in data collection and to pre-process leaf images that will be taken at small-scale crop farms in southern highlands of Tanzania. The two main crops are Common beans and Irish potatoes.

The four classes of Common beans are:

• Healthy (healthy)
• Bean rust (rust)
• Bean anthracnose (anthra)
• Other (other)

The four classes of Irish potatoes are:

• Healthy (healthy)
• Early blight (earlyblt)
• Late blight (lateblt)
• Other (other).

1. Data collection tool

The data collection tool is the Open Data Kit (ODK) app on Android smartphones. Data storage is on Google drive.

The ODK forms for collecting leaf images of Irish Potatoes and Common Beans are illustrated here below:

2. Data Pre-processing

The code files used for initial data pre-processing include:

• download_files.py: used Google Drive Downloader to download the crop leaf images from URLs captured by ODK. The leaf image is saved as a URL on Google sheet when the file is uploaded from the ODK form on the app. For example, a file with URL https://drive.google.com/open?id=1_vNgNDJxyh4QLe1vkoD0I34A-j8AU8jK is downloaded to a file named 1_vNgNDJxyh4QLe1vkoD0I34A-j8AU8jK.jpg

• rename_img.py: used to rename files in a given folder in the format classnameFilenumber.extension. The code renames a file with name 1_vNgNDJxyh4QLe1vkoD0I34A-j8AU8jK.jpg to healthy25.jpg. The code also produces a .csv file that tracks the new and old filenames. The format of the csv file is indicated below:

new_filename	old_filename
healthy1.jpg	1jR4L3QMFN_wGes5NSpECjf819i4BLhxC
healthy2.jpg	1zIcpXRMPUf-KcPwI53ohi1QIRdxR9Kc7
healthy3.jpg	1jEN6zBL_0RRMHanGRik54HK75GOhu1YF
.	.
.	.
.	.
healthy25.jpg	1_vNgNDJxyh4QLe1vkoD0I34A-j8AU8jK