The primary bottleneck for recruitment of fish populations is the early developmental period from fertilization of eggs to the first-feeding larval stage. During this period fish are very sensitive to pollutant exposure, and short-term exposure during embryonic development can result in deformed larvae and increased mortality. Such early life stage mortality and sub-lethal effects can influence the abundance and health of adult fish populations, including species important to commercial fisheries such as Atlantic cod. Monitoring the development of eggs and larval fish and understanding the effects of pollutant exposure can therefore improve future management of fish populations and inform risk analysis in industries such as oil exploitation and agriculture. Current in-situ monitoring methods rely on net sampling at sparse locations in the ocean, and labour-intensive manual imaging and measurement. We demonstrate a system for high-magnification, high-throughput imaging and analysis of fish eggs and larvae in the laboratory, and a similar system mounted on an AUV (Autonomous Underwater Vehicle) for in-situ imaging. Our systems are able to operate with little manual intervention and collect images of similar quality and magnification to traditional microscopy in a fraction of the time. We also show how we can use machine learning and classical machine vision with these images to automatically identify and measure eggs and larvae and obtain biologically relevant data. We present results of laboratory studies with pollutant-exposed fish as a proof of concept, and show how an AUV equipped with our system can collect images for analysis continuously over a large volume of water.