Abstract
Strawberries are an important and high-value crop in the United States. The total value of strawberry production in the United States was $3.42 billion in 2021. However, a major pest of strawberries, two-spotted spider mites (TSSM) (Tetranychus urticae Koch), can cause severe economic losses due to mites feeding on chlorophyll which interferes with photosynthesis. Predatory mites (Neoseiulus californicus and Phytoseiulus persimilis) can control TSSM populations and provide an alternative to conventional pesticides. Predatory mites are more friendly to the environment than pesticides. To monitor TSSM and predatory mite populations in the field, frequent sampling is required. However, manual counting of mites using a hand lens is challenging for a large field. Therefore, this study applied deep learning models to detect TSSM and predatory mites in strawberries from smartphone images. Three different smartphones (Motorola Moto E5 Play, OnePlus A5000, and iPhone XR) and a 25x macro lens were used to collect a total of 3,013 images, including 1,779 TSSM images and 1,234 predatory mite images. TSSM images were collected by the three smartphones, and only the OnePlus A5000 and iPhone XR were used to collect predatory mite images that only contained predatory mites or contained both predatory mites and TSSM in the same image. TSSM images were used to compare two deep learning models (YOLOv4 and Faster R-CNN Resnet50) and three different smartphone cameras. This study found that the average detection accuracy of YOLOv4 and Faster R-CNN was 0.935 and 0.863, respectively. Deep learning models achieved the best detection accuracy on the image collected by iPhone XR. The YOLOv4 model was trained to compare the detection performance of three different input image sizes (160 × 160, 320 × 320, 640 × 640 pixels) for TSSM and predatory mite detection. This study found that the mean average precision of the YOLOv4 increased as the image size increased, and the model trained with an image size of 640 × 640 pixels achieved the detection accuracy of 0.933 for TSSM and predatory mites. This study also found that the YOLOv4 could detect pests under different field illumination conditions. The smartphone image-based method can detect the TSSM and predatory mites quickly from images and with high detection accuracy, which can help growers use less time to monitor mite populations in the field.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.