CropGCNN: color space-based crop disease classification using group convolutional neural network.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Image classification is a key computer vision task, with deep, well-connected networks achieving top performance.
• Standard methods use Red Green Blue (RGB) images without exploring color space impact on accuracy.
• Datasets often have fixed color image formats, potentially limiting classification performance.

Purpose of the Study:

• To investigate the impact of various color spaces on image classification accuracy.
• To develop a model that leverages multiple color spaces simultaneously for improved classification.
• To reduce model complexity and computational load while maintaining high accuracy.

Main Methods:

• Input RGB images were transformed into seven distinct color spaces concurrently.
• Each color space was processed by a dedicated Convolutional Neural Network (CNN) model.
• Group convolutional layers were employed to reduce computational demands and hyperparameters.

Main Results:

• The multi-color space model demonstrated significant accuracy gains over state-of-the-art methods.
• Different image classes were found to be better represented in specific color spaces.
• The proposed model achieved high accuracy with substantially fewer parameters.

Conclusions:

• Utilizing multiple color spaces within a single model is crucial for enhancing image classification.
• This approach offers a more efficient and accurate method for complex datasets like crop disease identification.
• The findings suggest a new direction for designing robust and resource-efficient deep learning models.