Skip to content

dataset-overview.md

Latest commit

 

History

History
86 lines (57 loc) · 2.06 KB

dataset-overview.md

File metadata and controls

86 lines (57 loc) · 2.06 KB
Raw

Dataset Overview

Project

AI-Powered Weed Detection System for Smart Agriculture

Objective

The objective of this project was to develop a computer vision model capable of detecting weeds within bean farms to support precision agriculture and reduce manual weed identification efforts.

Dataset Summary

Metric Value
Total Images 213
Total Annotations 2334
Classes 2
Class Names Bean, Weed
Average Annotations Per Image 9.3
Image Resolution 3024 × 4032 (median)

Dataset Collection

Images were collected from agricultural environments containing bean crops and naturally occurring weeds. The dataset was designed to simulate real-world farm conditions.

Annotation Process

All images were manually annotated using Roboflow.

Classes:

Bean

  • Total annotations: 296

Weed

  • Total annotations: 2038

Bounding boxes were used to identify and localize objects within each image.

Dataset Split

Split Images Percentage
Train 1913 82%
Validation 256 11%
Test 163 7%

Preprocessing

The following preprocessing steps were applied:

  • Auto Orientation
  • Static Crop
  • Resize to 640x640
  • Contrast Stretching

Data Augmentation

The following augmentations were applied:

  • Horizontal Flip
  • 90 Degree Rotation
  • Zoom Augmentation
  • Blur
  • Motion Blur
  • Noise Injection

These augmentations were used to improve model generalization and robustness.

Challenges

  • Small dataset size
  • Class imbalance between bean and weed classes
  • Variations in lighting conditions
  • Differences in plant growth stages

Skills Demonstrated

  • Dataset Engineering
  • Computer Vision
  • Data Annotation
  • Data Preparation
  • Machine Learning Workflows
  • Agricultural AI