Skip to content
This repository was archived by the owner on Jun 24, 2024. It is now read-only.

Latest commit

 

History

History
61 lines (49 loc) · 3.19 KB

USECASE.md

File metadata and controls

61 lines (49 loc) · 3.19 KB

Machine Learning Canvas

Background

  • users: We target farmers, botanists, agricultural scientists, environmentalists, nature enthusiasts.
  • goals: Enable our users to determine whether a leaf is sick and identify the specific disease it has without any prior botanical knowledge.
  • pains: Determining the health status of a leaf often requires botanical knowledge, which might be time-consuming to acquire.

Value Proposition

  • product: We propose an application for automatically identifying the health status and potential disease of a leaf based on a picture.
  • alleviates: Users do not need to have any specific botanical knowledge.
  • advantages: Users have an accurate and straightfoward way of identifying the health status and potential disease of a leaf.

Objectives

  1. Provide the user with a responsible API and/or interface to use our product
  2. Classify a leaf based on a picture of it, using the model that is the most suited to the user's needs

Solution

  • core features: - Predict the specific illness afflicting the leaf. - User feedback process for incorrectly classified illness.
  • integration: - The model will run on a VPS as well as its API to interact with it. - The API will be used to create a user-friendly web interface for uploading and predicting on user images.
  • alternatives: - Allow users to add content manually and classify them.
  • constraints: - Maintain low latency when predicting user inputs. - Complete category of illnesses so that the model can predict any of them.
  • out-of-scope: - More sophisticated model to increase the accuracy (More: ram usage & inference latency).

Feasibility

  • data: We have a dataset composed of ~40,000 annotated images
  • team: We are three experienced data scientists
  • infrastructure:
    • local: We will use OVH hosting services to deploy our API and serve our model (using Flask) by ourselves.
    • cloud: We will deploy our model in the cloud as well.

Data

  • Kaggle dataset composed of ~40,000 annotated leaves images with 63 different leaf disease classes. We will devide the data into trainnig and testing sets with 80% and 20% respectively for each class we have.

Metrics

  • Binary precision and recall, i.e., healthy from unhealthy leaf detection.
  • Per-class precision and recall.
  • Our focus will be on recall and especially for binary classification since we beleive that the cost of missing a sick leaf is very high.

Evaluation

  • ?

Modeling

  • Deep learning model using convolutional neural networks.

Inference

  • We opt for online inference method, i.e. handling real-time requests, where a user can interact with our API and/or interface online.

Feedback

  • Allow users to report issues related to misclassification.
  • Allow users to upload images with their corresponding illnesses.

Project