Comparison of Plant Detection Performance of CNN-based Single-stage and Two-stage Models for Precision Agriculture

The fact that arable land is not increasing in proportion to the ever-increasing population will increase need for food coming years. For this reason, it necessary yield of crops make optimum use land. One most important reasons decrease and quality weeds. Herbicides are generally preferred weed management. Due deficiencies herbicide application methods, only 0.015-6% herbicides reach their target. herbicides, which an part agricultural system, issue needs be emphasized, considering risk residue environmental damage. In parallel with rapid development electronic computer technologies, artificial intelligence applications have had opportunity develop. context, plant detection subsystems machines contribute precision agriculture techniques. study, performances single-stage two-stage Convolutional Neural Network (CNN)-based deep learning (DL) models evaluated. a dataset was created by taking images Zea mays, Rhaponticum repens (L.) Hidalgo, Chenopodium album L. plants lands Konya. With dataset, training carried out transfer method. evaluation metrics trained were calculated using error matrix. addition, time prediction used as quantitative models. performance, time, 85%, 8 h, 1.21 s SSD MobileNet v2 99%, 22 2.32 Faster R-CNN Inception v2, respectively. According these results, outperform terms accuracy. However, cases where important, model can more data its