No abstract

Automatic detection of foliar diseases in potato fields, such as early blight caused by Alternaria solani, could allow farmers to reduce the application of plant protection products while minimizing production losses. UAV-based, high resolution, NIR-sensitive cameras offer the advantage of a detailed top-down perspective, with high-contrast images ideally suited for detecting Alternaria solani lesions. A field experiment was conducted with 8 plots housing 256 infected plants which were monitored 6 times over a 16-day period with a UAV. A modified RGB camera, sensitive to NIR, was combined with a superzoom lens to obtain ultra-high-resolution images with a spatial resolution of 0.3 mm/px. More than 15,000 lesions were annotated with points in two full size images corresponding to 1250 cropped tiles of 256 by 256 pixels. A deep learning U-Net model was trained to predict the density of Alternaria solani lesions for every pixel. In this way, density maps were calculated to indicate disease hotspots as a guide for the farmer.

Potato cultivation is regularly affected by Alternaria solani, a destructive foliar pathogen causing early blight, a premature defoliation of potato plants resulting in yield losses. Currently, Alternaria is treated through preventive application of chemical crop protection productions, following warnings based on weather predictions and visual observations. Automatic detection could make the mapping of early blight more accurate, reducing production losses and application of crop protection products. Current research explores the potential of deep learning of high resolution imagery within precision agriculture, mainly using supervised learning. However, available datasets are often limited in size and variation, which reduces the robustness of the developed models. Here, we present a convolutional network to detect Alternaria and evaluate the influence of sampling size, sampling balance and sampling accuracy on the model performance. These analyses are based on ultra-high-resolution datasets of modified RGB cameras obtained with unmanned aerial vehicles (UAV) and collected over experimental in-field Alternaria trials. By using this varied dataset instead of a single-time dataset, higher accuracies are achieved.The method is relatively robust for imbalances of the training dataset. Further, we show that labeling quality plays a role, but that an error of up of to 20% of labeling is acceptable for good results. In conclusion, extra variability leads to more robust disease detection, desirable for in-field application.