Automatic Tomato Seedling Segmentation for Genotypic Assessment via Gaussian Mixture Models

Tomatoes are one of the most iconic crops in the Southern Mediterranean area, playing an important role both in social and economic terms, mainly due to large parts of the population strictly relying on these crops as the main food and income source. Consequently, variations in the yields of tomatoes could have large social and economic implications. However, extreme weather events, such as droughts, are increasingly common and widespread due to the ongoing climate crisis. Therefore, assessing the behaviour of crops in response to this type of stress is extremely important to guarantee a stable and reliable food source. To this end, this work proposes an end-to-end tomato evaluation strategy based on image processing and machine learning algorithms. The proposed pipeline allows domain experts to automatically evaluate relevant phenotypical traits of a population of genotypic mutants, comparing them with their Red Setter regarding the response to drought stress. The framework uses a simple, low-cost, and easy-to-use setup, which allows the rapid acquisition of a wide dataset. This can be acquired in different instants, providing temporal-dependent information on the response of the crops to different scenarios, which can be used to estimate relevant phenotypical traits. To evaluate its effectiveness, the processing pipeline was evaluated over two weeks, with a total of 388 mutants compared with 12 Red Setter, exploiting canopy coverage to assess the overall resnonse of the mutants to drought stress.