Analytica chimica acta · 2026
Dual-modal deep learning model with microstrip isoelectric focusing and imaging strategy for feed classification.
СтатьяПоддержка исследования: вне правительства СШАНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Tian Youli, Lu Xiunan, Cao Yiren, Chen Ke-Er, Zhang Qiang, Liu Weiwen, Sun Chunhua, Xu Xu, Wang Yuxing, Cao Chengxi
Аннотация
BACKGROUND: Accurate identification of animal-derived feed ingredients is essential for ensuring livestock safety and preventing disease transmission. However, conventional analytical methods (e.g., PCR, microscopy) are often hindered by high costs, low throughput, or limited universality. Furthermore, single-modal approaches face inherent "blind spots": biochemical methods may struggle with biologically homologous species, while visual inspection fails with indistinguishable powders. Therefore, there is a critical need for an intelligent analytical strategy that synergizes multidimensional information for robust classification in complex matrices.
RESULTS: Herein, we established a dual-modal deep learning framework integrated with an array-format microstrip isoelectric focusing (mIEF) device. This platform simultaneously captures orthogonal analytical features: microscopic biochemical fingerprints via stable mIEF profiling, and macroscopic physical textures via digital RGB imaging. To maximize identification accuracy, a dual-branch ResNet18 model equipped with a channel attention mechanism was constructed. This algorithm adaptively fuses the two modalities, effectively compensating for the limitations of individual sensors. The proposed model achieved a superior classification accuracy of 96.53% across 15 feed categories, significantly outperforming single-modal baselines (mIEF-only: 83.47%; image-only: 49.33%). Notably, the integrated device supports parallel processing of 12-24 samples per run at an estimated cost of less than $1.5 per sample.
SIGNIFICANCE: This study presents a practical analytical strategy for feed traceability by integrating mIEF-based protein fingerprinting with digital image analysis. The combination of biochemical and visual information improves classification performance compared with either modality alone, while the array-format device supports parallel sample processing at low cost. The proposed platform provides a feasible approach for routine screening of animal-derived feed ingredients, particularly in laboratories where rapid, accessible, and cost-effective methods are needed.
RESULTS: Herein, we established a dual-modal deep learning framework integrated with an array-format microstrip isoelectric focusing (mIEF) device. This platform simultaneously captures orthogonal analytical features: microscopic biochemical fingerprints via stable mIEF profiling, and macroscopic physical textures via digital RGB imaging. To maximize identification accuracy, a dual-branch ResNet18 model equipped with a channel attention mechanism was constructed. This algorithm adaptively fuses the two modalities, effectively compensating for the limitations of individual sensors. The proposed model achieved a superior classification accuracy of 96.53% across 15 feed categories, significantly outperforming single-modal baselines (mIEF-only: 83.47%; image-only: 49.33%). Notably, the integrated device supports parallel processing of 12-24 samples per run at an estimated cost of less than $1.5 per sample.
SIGNIFICANCE: This study presents a practical analytical strategy for feed traceability by integrating mIEF-based protein fingerprinting with digital image analysis. The combination of biochemical and visual information improves classification performance compared with either modality alone, while the array-format device supports parallel sample processing at low cost. The proposed platform provides a feasible approach for routine screening of animal-derived feed ingredients, particularly in laboratories where rapid, accessible, and cost-effective methods are needed.