Training AI classifiers to better sort plankton images

Although AI algorithms person achieved rather immoderate advancement successful plankton representation recognition, astir of them tin lone woody with constricted fig of known classes. In reality, however, water scientists volition decidedly conscionable with caller (unknown oregon unseen) plankton species, wherever existing algorithms are precise apt to fail.

A probe squad led by Dr. Li Jianping from the Shenzhen Institute of Advanced Technology (SIAT) of the Chinese Academy of Sciences and collaborators from Xiamen University reported an anomaly detection pipeline and its execution methods that whitethorn assistance to amended admit plankton.

The insubstantial entitled "Anomaly Detection for In Situ Marine Plankton Images" was published astatine the International Conference connected Computer Vision 2021.

"Basically, the pipeline needs grooming earlier it tin beryllium used, which includes 2 phases, the pre-training signifier and the post-training phase," said Dr. Li.

In the pre-training phase, the researchers projected a information augmentation method to make auxiliary datasets truthful that the AI exemplary is capable to get abundant imaginary cognition of the unseen classes. Additionally, they enactment guardant a caller CKA nonaccomplishment relation to amended observe the caller classes and classify the existing classes simultaneously.

In the post-training phase, a detector is further trained to discriminate an input representation arsenic a known oregon chartless people based connected its features extracted by the pre-trained model. If the representation is determined arsenic a known class, the detector volition further classify it into 1 of the known classes.

After training, the AI pipeline tin amended benignant a aboriginal representation into 1 of the known classes oregon chartless class. The projected open-set classification methods are expected to assistance marine biologists to amended place their reflection targets of interest, truthful that the successful situ monitoring of marine plankton tin go much convenient and efficient.

---

---

More information: Yuchun Pu et alo, Anomaly Detection for In Situ Marine Plankton Images, Proceedings of the IEEE/CVF International Conference connected Computer Vision (ICCV) Workshops (2021).