Investigating the abundance, distribution and diversity of plankton organisms holds a key element in understanding complex, marine food web structures, nutrient cycling, climate change effects and anthropological influences on the world’s largest habitat. Over the past years, plankton research has developed from discrete, integrating net samples to fine-scale, in situ, optical detection of organisms. Camera systems like the Lightframe On-Sight Keyspecies Investigation (LOKI) system allow a high resolution in capturing organisms below the size of 60 µm, while connecting the image data with environmental measurements. In this thesis the processing of LOKI plankton images, recorded 2014 in the Sognefjord, Norway, is investigated and improved from edge detection over feature selection to the classification of morphological groups with multivariate and machine learning classifiers. For the classifications, a subset of calculated image features is determined individually for each method by evaluating a decreasing number of image feature ranked by their Gini Index significance. Transfer learning is used to implement a fine-tuned AlexNet convolutional neural network (CNN) as a first approach towards deep learning classification of LOKI images. The improvement of the image processing and edge detection was successful, as the implementation of a Canny edge detection algorithm detects the edges much closer to the organism. With the selected features the highest classification accuracy of 88.72 % is achieved with a random Forest classification while the transfer learning CNN results achieve a classification accuracy of 87.75 %. This thesis has laid out new approaches for LOKI plankton image classification, which will help to progress providing a complete processing chain from image capturing towards autonomous classification and presorting of major morphological groups.