JASS-Net: Joint Attention-selection Mapping for Local Coconut Seedling Classification

Read full text Summarise with VIRAT
Authors: Muh. Rasyid Ridha, Dwi Yuli Prasetyo, Abdul Muni
Cited by: 0
References: 52

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

JASS-Net is a proposed model for the classification of local coconut seedlings, designed to address the challenges of fine-grained classification with high computational efficiency.The model integrates a Lightweight CNN Backbone with three key modules: Lightweight Attention Extractor (LAE), Adaptive Feature Selector (AFS), and Joint Attention-Selection Mapping (JASM), enabling the extraction and interaction of relevant features for precise classification.Evaluation on an independent test set demonstrated outstanding performance, achieving an overall accuracy of 98.01%, Macro Precision, Recall, and F1-score of 98.01%, Top-2 Accuracy of 99.62%, and Top-3 Accuracy of 99.89%.All classes attained F1-scores above 0.96, with the "Immature" class achieving the highest F1 of 0.999, reflecting robust detection of subtle distinguishing features.JASS-Net also exhibits high computational efficiency, with a model size of 0.47 MB, 122,949 parameters, and an inference latency of 0.001346 s per sample, making it suitable for deployment on resource-limited devices.These results highlight JASS-Net ability to perform accurate, balanced, and efficient fine-grained classification, with potential applicability to other plant species and future improvements through transfer learning and larger datasets.

Fields of Study & Subjects

Smart Agriculture and AI Advanced Neural Network Applications Domain Adaptation and Few-Shot Learning

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
812–829
Publication date
23 March 2026
Language
EN