Hair Fescue and Sheep Sorrel Identification Using Deep Learning in Wild Blueberry Production

Deep learning convolutional neural networks (CNNs) are an emerging technology that provide opportunity to increase agricultural efficiency through remote sensing and automatic inferencing of field conditions. This paper examined the novel use CNNs identify two weeds, hair fescue sheep sorrel, in images wild blueberry fields. Commercial herbicide sprayers a uniform application agrochemicals manage patches these weeds. Three object-detection three image-classification were trained sorrel using from 58 The 1280x720 tested at four different internal resolutions. retrained with progressively smaller training datasets ranging 3780 472 determine effect dataset size on accuracy. YOLOv3-Tiny was best CNN, detecting least one target weed per image F1-scores 0.97 for 0.90 1280 × 736 resolution. Darknet Reference most accurate classifying containing 0.96 0.95, respectively 736. MobileNetV2 achieved comparable results lowest resolution, 864 480, 0.95 both Training had minimal accuracy all except Reference. can be used smart sprayer control specific spray applications, reducing use. Future work will involve testing development growers field-specific information. Using improve create major cost-savings producers.