Azure UNDP Weed Detection

Datasets

Data being used in this project is V2: Nonsegmented single plants (1.7GB) from PAPER: A Public Image Database for Benchmark of Plant Seedling Classification Algorithms.

It contains about 960 unique plants of 12 classes:

└── NonsegmentedV2
    ├── Black-grass
    ├── Charlock
    ├── Cleavers
    ├── Common Chickweed
    ├── Common wheat
    ├── Fat Hen
    ├── Loose Silky-bent
    ├── Maize
    ├── Scentless Mayweed
    ├── Shepherd’s Purse
    ├── Small-flowered Cranesbill
    └── Sugar beet

More information can be found from Plant Seedlings Dataset.

Training Weed Detection Model

The details can be found in notebooks/UNDP.ipynb and can be run with Colab.

Azure Architectures

Azure Functions

Azure Notification Hubs

Requirements:

• Azure CLI
• Docker
• Microsoft Azure Account
• Docker Hub Account

Azure Notification Hub Setup

1. Configure Azure Notification Hub and connect app to the notification hub. (See Tutorial: Send push notifications to Android devices using Firebase SDK version 0.6).
2. Replace <ConnectionString> and <HubName> in undp/classify/__init__.py with DefaultFullSharedAccessSignature and name of the hub you have configured.

Docker Setup

cr. fastai

1. Open your terminal and change directory to undp
2. Build Docker image

   docker build --tag <DOCKER_HUB_ID>/<DOCKER_IMAGE_NAME>: .

3. Test Docker image

   docker run -p 8080:80 -it <DOCKER_HUB_ID>/<DOCKER_IMAGE_NAME>:

4. Open browser:

   localhost:8080/api/classify?name=https://raw.githubusercontent.com/nithiroj/azure-undp-weed-detection/master/assets/maize-251.png

5. Push the image to Docker Hub

   docker login

   docker push <DOCKER_HUB_ID>/<DOCKER_IMAGE_NAME>:

Azure Functions Setup

cr. fastai

1. Login to Microsoft Azure with Azure CLI

   az login

2. Run following command to see list of available location for <LOCATION_ID>, e.g. centralus.
3. Create resource group

   az group create \
   --name <RESOURCE_GROUP> \
   --location <LOCATION_ID>

4. Create storage account

   az storage account create \
   --name <STORAGE_ACCOUNT> \
   --location <LOCATION_ID> \
   --resource-group <RESOURCE_GROUP> \
   --sku Standard_LRS

5. Create a Linux App Service Plan

   az appservice plan create \
   --name <APP_PLAN_NAME> \
   --resource-group <RESOURCE_GROUP> \
   --sku B1 \
   --is-linux

6. Create the App & Deploy the Docker image from Docker Hub

   az functionapp create \
   --resource-group <RESOURCE_GROUP> \
   --name <FUNCTION_APP> \
   --storage-account <STORAGE_ACCOUNT> \
   --plan <APP_PLAN_NAME> \
   --deployment-container-image-name <DOCKER_HUB_ID>/<DOCKER_IMAGE_NAME>:

7. Configure the function app

   storageConnectionString=$(az storage account show-connection-string \
   --resource-group <RESOURCE_GROUP> \
   --name <STORAGE_ACCOUNT> \
   --query connectionString --output tsv) az functionapp config appsettings set --name <FUNCTION_APP> \
   --resource-group <RESOURCE_GROUP> \
   --settings AzureWebJobsDashboard=$storageConnectionString \
   AzureWebJobsStorage=$storageConnectionString

8. Run your Azure Function

   https://<FUNCTION_APP>.azurewebsites.net/api/classify?name=<IMAGE_URL>

Delete Resource Group

When you are done, delete the resource group

az group delete \
--name <RESOURCE_GROUP> \
--yes

Demo

The demo can be tested by replacing <IMAGE_URL> with your image url.

https://undpinferenceapp.azurewebsites.net/api/classify?name=<IMAGE_URL>

For example:

https://undpinferenceapp.azurewebsites.net/api/classify?name=https://raw.githubusercontent.com/nithiroj/azure-undp-weed-detection/master/assets/maize-251.png

Note:

• More samples of images are avilable in assets/
• The demo will be available until the end of June 2020.