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.Artificial intelligence technologies, including machine learning and deep learning, has shown the potential of transforming agriculture beyond our imaginations thanks to their abilities in extraction of hidden-layer information of big data. In recent years, deep learning algorithms have been widely studied to process and analyze imagery data collected using UAV-based imaging systems in precision agriculture and plant high throughput phenotyping. Deep-learning based data analytic methods acquired higher accuracy in estimation of crop yield, plant height, disease infection and weed detection. The goal of this paper was to predict the cotton yield using an improved Long short-term memory (LSTM) model, which is an artificial recurrent neural network (RNN) architecture used in the field of deep learning. LSTM model has feedback connections and combinations of different gates (such as input gate, forget gate and output gate) to control the information needed to be memory for the previous time stamp data and to be updated from this time stamp inputs. In this study, a UAV imaging system consisting of a multispectral camera of five narrow spectral bands of blue (475±10 nm), green (560±10 nm), red (668±5 nm), red-edge (717±5 nm) and near-infrared (840±20 nm) was used to collect imagery data of cotton in three critical growth stages. Imagery data were pre-processed to remove background, calibrate reflectance and register to yield data based geo-referenced information. Multivariable factors of NDVI, GNDVI, canopy size and canopy temperature were extracted from UAV multispectral images and used as the input for the LSTM model. The parameters of the LSTM model with be optimized to improve the performance for accurate yield estimation.