计算机科学
人工智能
深度学习
浮游植物
可靠性(半导体)
灵活性(工程)
过程(计算)
人工神经网络
特征提取
机器学习
模式识别(心理学)
生态学
数学
生物
功率(物理)
统计
物理
量子力学
营养物
操作系统
作者
David Rivas-Villar,José Morano,José Rouco,Manuel G. Penedo,Jorge Novo
标识
DOI:10.1007/978-3-031-25312-6_49
摘要
Certain phytoplankton species can produce potent toxins that can raise health concerns, especially if these species proliferate in water sources. Furthermore, phytoplankton can drastically and rapidly multiply their population in water, increasing the possibility of dangerous contamination. Nowadays, preventive water analyses related to phytoplankton are routinely and manually performed by the experts. These methods have major limitations in terms of reliability and repeatability as well as throughput due to their complexity and length. Therefore, the automatization of these tasks is particularly desirable to lower the workload of the experts and ease the whole process of potability analysis. Previous state of the art works can segment and classify phytoplankton from conventional microscopy images of multiple specimens. However, they require classical image features, which need ad-hoc feature engineering, a complex and lengthy process. Thus, employing novel deep learning-based deep features is highly desirable as it would improve the flexibility of the methods. In this manuscript, we present a study regarding the performance of different pre-trained deep neural networks for the extraction of deep features in order to identify and classify phytoplankton. The experimental results are satisfactory as we improve the performance of the state of the art approaches. Furthermore, as we eliminate the need for classical image features, we improve the adaptability of the methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI