DPXception: a lightweight CNN for image-based date palm species classification.

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Technology

Background:

• Date palm species classification is crucial for agriculture and economics.
• Existing methods often rely on fruit traits, limiting their applicability.
• Image-based classification presents a viable alternative but requires robust models and datasets.

Purpose of the Study:

• To introduce a novel dataset for date palm species classification.
• To develop an efficient and accurate image-based classification model.
• To create a practical application for real-time species identification.

Main Methods:

• Collected and augmented a dataset of 2358 images across four date palm species.
• Developed DPXception, a lightweight CNN model adapted from Xception, using the first 100 layers.
• Implemented normalization and global average pooling to optimize the model.

Main Results:

• DPXception achieved 92.9% accuracy and 93% F1-score, surpassing seven benchmark models.
• The model demonstrated the lowest inference time at 0.0513 seconds.
• Developed an Android application for real-time, on-device date palm species classification.

Conclusions:

• The developed dataset and DPXception model offer a robust solution for image-based date palm species classification.
• This work provides the first public dataset and a practical, efficient classification method.
• Future research can extend this to date palm gender classification and age estimation.