An Attention-based Deep Learning Model for Jasminum Quality Grading

Jasmine flowers, which are much cultivated in the states of Tamil Nadu and Karnataka, contribute more than ₹20 crores annually to the Indian economy through internal sales and exports. Traditional grading practices, however, suffer from human subjectivity, often resulting in inconsistencies. This paper presents a deep-learning-based grading system that utilizes the InceptionResNetV2 model with self-attention mechanisms to classify flowers based on bud tip size, freshness, color, and tubule presence. The system focuses on fine-grained details rather than the whole flower and achieves 92% accuracy, which ensures consistency and fairness in quality assessment. The automated approach is significantly more efficient, transparent, and scalable and likely to be compatible with global trade standards of jasmine flowers.