Microstructure segmentation with deep learning encoders pre-trained on a large microscopy dataset

This study examined the improvement of microscopy segmentation intersection over union accuracy by transfer learning from a large dataset of microscopy images called MicroNet.Many neural network encoder architectures were trained on over 100,000 labeled microscopy images from 54 material classes.These pre-trained encoders were then embedded into multiple segmentation architectures including UNet and DeepLabV3+to evaluate segmentation performance on created benchmark microscopy datasets.Compared to ImageNet pre-training,models pre-trained on MicroNet generalized better to out-of-distribution micrographs taken under different imaging and sample conditions and were more accurate with less training data.When training with only a single Ni-superalloy image,pre-training on MicroNet produced a 72.2%reduction in relative intersection over union error.These results suggest that transfer learning from large in-domain datasets generate models with learned feature representations that are more useful for downstream tasks and will likely improve any microscopy image analysis technique that can leverage pre-trained encoders.

Optimising mesh size with escape gaps in a dual-species portunid-trap fishery

In south-eastern Australia,the same baited,round traps(comprising 50–57-mm mesh netting)are used to target giant mud,Scylla serrata and blue swimmer crabs,Portunus armatus in spatially separated fisheries.Both fisheries are characterised by the common,problematic discarding of undersized portunids(<85 and 65 mm carapace length;CL for S.serrata and P.armatus)and fish(yellowfin bream,Acanthopagrus australis).This poor selectivity was addressed here in two experiments assessing the utility of(1)traps partially or completely covered in larger mesh(91 mm to match the minimum legal size of the smaller P.armatus),and then(2)any cumulative benefits of fitting species-specific escape gaps.In experiment 1,there were no differences among catches of legal-sized portunids associated with either partial,or complete trap coverage with larger mesh.Irrespective of mesh coverage,both designs of 91-mm traps also retained significantly fewer(by up to 42%)undersized P.armatus and A.australis.In experiment 2,replicate traps completely covered in 91-mm mesh were tested against conventional traps comprising 56-mm mesh,and traps with the same mesh sizes,but also three escape gaps configured for either S.serrata(46×120 mm)or P.armatus(36×120 mm)(i.e.four treatments in total).All modified traps maintained catches of legal-sized S.serrata,and only the 91-mm traps with escape gaps caught fewer legal-sized P.armatus.Fewer undersized S.serrata,P.armatus and A.australis(mean catches reduced by up to 49%)were retained in all larger-meshed than small-meshed traps,and in all of those traps with escape gaps(by up to 95%)than without.While there were no significant cumulative benefits of escape gaps in largermeshed traps(measured by a statistical interaction),there was a trend of fewer unwanted catches overall.These data support configuring portunid traps with mesh sizes matching the morphology of the smallest legal-sized target species.But,simply retroactively fitting escape gaps in existing,smaller-meshed traps will also realize positive selectivity benefits.