Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract.

Cucumber (Cucumis Sativus L.) is an important crop popular for consumers but vulnerable to many diseases. Conventionally, cucumber diseases are identified mainly by human based on observable foliar symptoms. Sometimes, one plant could be invaded by multiple diseases at the same time. This would cause complicated patterns and poses a significant challenge for identification. To improve the situation, the development of an automatic identification system is essential. This study proposed an identification model that could identify seven cucumber foliar diseases with their coexistence cases. One-hot and multi-hot labeling were used for annotation. A total of 8810 cucumber leaf images were collected from greenhouses. The disease categories covered anthracnose, corynespora blight, downy mildew, powdery mildew, virus, leaf miner, and malnutrition. A faster region-based convolution neural network (Faster R-CNN) with a ResNeXt-101 backbone was trained to identify the diseases. A resampling method was applied to mitigate the deficiencies numbers of images in certain categories. In addition, a modified cross entropy loss, named concurrent cross entropy, was applied to enhance the training of multi-disease. The trained model achieved an F₁-score of 0.846, a mean average precision (mAP) of 0.758, and an accuracy of 0.733. Future studies will aim to evaluate the disease severity and further improve the overall identification performance.