Weeds detection efficiency through different convolutional neural networks technology

The preservation of the environment has become a priority and subject that is receiving more attention. This particularly important in field precision agriculture, where pesticide herbicide use controlled. In this study, we propose to evaluate ability deep learning (DL) convolutional neural network (CNNs) technology detect weeds several types crops using perspective proximity images enable localized ultra-localized spraying region Beni Mellal Morocco. We studied detection through six recent CNN known for their speed precision, namely, VGGNet (16 19), GoogLeNet (Inception V3 V4) MobileNet (V1 V2). first experiment was performed with CNNs architectures from scratch second pre-trained versions. results showed Inception V4 achieved highest rate 99.41% 99.51% on mixed image sets its version respectively, V2 fastest lightest size 14 MB.