I think this should be possible with ImageJ using classification methods which are available as ImageJ plugins.
A frequent used plugin is, e.g., the Trainable Segmentation plugin:
A simple workflow would be to identify and segment the different classes (organisms) and then count them with the particle analysis of ImageJ, see:
You can easily apply those methods on a folder and subfolder of images.
Another option for the classification would be to use R within Bio7 (which integrates ImageJ as a plugin). I've created a video tutorial how to train a R supervised classifier which then could also be used on a folder of images:
There seems to be also a special R project available (using ImageJ) to classifiy zooplankton in a similar way: