0

Soybean Seed Classification

Django | Python | Computer Vision Engineer | Image Processing | Deep Learning and Neural Networks | Filtering

Introduction

Soybean Seed Classification | Computer Vision

  • Led the development of a computer vision system for soybean seed classification.

  • Collected diverse seed dataset and

  • applied advanced image processing for enhancement

  • Evaluated and selected ResNet50 for optimal seed classification.

  • Implemented user-friendly web-based UI for real-time results and defect identification.

  • Demonstrates expertise in computer vision, image processing, and deep learning.

    Download PPT

Table of Content

Data Preparation and Preprocessing:

1. Soybean Image Dataset:

I assembled a comprehensive dataset comprising images of both healthy and defective soybean seeds.

2. Image Segmentation:

Utilized image segmentation techniques to distinguish soybean seeds from the background, enhancing the precision of subsequent analyses.

3. Contrast Detection:

Implemented contrast detection algorithms to identify variations in intensity, improving the seed's visibility.

4. Contour Enhancement:

Applied contour enhancement methods to emphasize the structural details of soybean seeds.

Neural Network Model Training:

Performed extensive training on various neural network models using the preprocessed data:

  • Explored models such as Inception V3, MobileNet, ResNet50, Sequential Scratch.
  • Based on performance metrics, ResNet50 was selected as the optimal neural network model.

Testing the Model:

1. Image Captured:

Introduced new images of soybean seeds to evaluate the model's generalization.

2. Image Segmentation:

Applied segmentation techniques to the test images for precise identification of seed boundaries.

3. Contour Detection:

Utilized contour detection algorithms to highlight seed features in the test images.

4. Contrast Enhancement:

Enhanced image contrast to ensure clarity and accurate analysis of seed characteristics.

Final Classification:

Implemented the trained ResNet50 model to classify soybean seeds into distinct categories:

  • Normal Seed
  • Purple Seed
  • Green Seed
  • Wrinkled Seed
  • Other Seed

Web-UI Development:

Developed a user-friendly web interface enabling users to upload soybean seed images and receive instant results regarding seed quality. The interface not only determines whether the seed is good or defective but also specifies the type of defect present.

Image Filtering Selection:

After thorough experimentation with various filter combinations (e.g., GBMS, GBSM, GMBS), the SMBG combination was chosen for its superior Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) values.

This comprehensive approach ensures accurate soybean seed classification and defect identification, making it a robust solution for agricultural applications.