Jungin Kim, Chan-young Yoon, Kwnag-Eun Ko, Inhoon Jang, "A Study on Adjusting Feature Vectors in Latent Space Using the Triplet Loss Function", International Journal of Control, Automation, and Systems, vol. 23, no.5, pp 1497-1509, 2025

The agriculture and fisheries industry faces an increasing need to transition from labor-intensive, experience-based methods to a data-driven smart industry, enhancing productivity through automation and un- manned systems. This paper proposes a modified ConvNeXt2 CNN model that utilizes the triplet loss function, tailored for agricultural and marine products requiring quality standard adjustments. This approach addresses the limitations of traditional classification models that rely on softmax and cross-entropy by enabling the direct handling of feature vectors in the latent space, allowing quality criteria adjustments without retraining the model. The pro- posed method demonstrated superior classification performance on actual agricultural and marine product datasets and confirmed the feasibility of arranging individual items by quality grades, facilitating immediate quality standard adjustments in response to market conditions.

A_Study_on_Adjusting_Feature_Vectors_in_Latent_Space_Using_the_Triplet_Loss_Function_IJCAS_논문.pdf