Disease and stress detection in strawberries by means of visual computing

Around 130,000 tonnes of strawberries are harvested in Germany every year. The infestation of strawberry plants by viruses, bacteria and fungi is the cause of considerable unwelcome costs, firstly due to loss of production and secondly due to increased expenditure on pesticides and other plant protection agents. Among the fungi, powdery mildew (Sphaerotheca macularis), gray mold (Botrytis cinerea) and root rot (Phytophthora spec.) are the most prevalent pathogens. Early detection of fungal diseases can prevent the spread of infection and save growers wasted investment in spoilt produce and additional costs for pesticides. 

The symptoms of an infestation are often only recognized late, when the infection is already well advanced and larger stands of crops are already infected. This can all be avoided by the evaluation of data supplied by a multispectral or hyperspectral camera, which also provides information in the infrared range. In analyzing this data, machine learning systems are capable of finding correlations between optical characteristics of the foliage and an infection of the root – objectively, speedily and reproducibly.

A practical field test has been set up in the framework of this industrial project to investigate the fundamental potential of multi- and hyperspectral sensor technology for the early detection of disease and pest infestation on strawberry farms. Mounted on a semi-mobile trolley that travels along overhead rails in a polytunnel, the camera records images at regular intervals with centimeter-precise localization in the visible and near-infrared range at multi- and hyperspectral resolutions. These images together with the disease and infestation rating data they deliver serve as training material for AI algorithms, which recognize and pinpoint the actual diseases while at the same time detecting the biotic stress due to the infection pressure at an early stage, in particular by taking into account the hyperspectral information. 

A productive combination of spectral information and machine learning facilitates the speedy, non-invasive and cost-effective examination of a strawberry crop for fungal infestation. Time series analyses with hyperspectral cameras also enable early recognition of infection onset, thereby allowing prompt intervention, damage prevention and reduced recourse to pesticides, with ultimate cost savings and mitigation of environmental impact.  

Theoretically, this technology can also be transferred to other crops grown in polytunnels. Greenhouse conditions offer scope for fixed-installation monitoring set-ups such as camera trolleys or for flexible systems such as robots to analyze plant health as well as harvesting and cultivation methods and thus to detect and pinpoint disease hotspots or pest infestation before it has taken hold.