Machine Learning for Plant Phenotyping and Seed Testing
Machine learning is a powerful tool that largely improves classification of biological objects compared to traditional image processing. Thereby, labelling of target objects in images is used to train the algorithms in finding the desired features in images. In complex rating tasks, classical image processing can fail because of similarity among different target objects, or low contrast to the background.
Briefly summarised the process works like this: in an image of germinating seeds (here wheat) shoots, roots, seeds, and background were labelled manually and the labelled image (blue border), together with the original image (grey border) were loaded into the machine learning algorithm.
In a pixel-based learning process features were trained to the algorithm and subsequently, the algorithm was able to find the features back in the image (green border).
After the training, the algorithm can apply the same feature extraction to other images that were not used for the training process.
Machine learning is available for all LemnaTec phenotyping and seed testing technology and can be set up for user-specific application cases.