Towards 3D CNN for precise crop yield estimation using multimodal remote sensing data: Case study of wheat in Morocco

Abstract

Crop yield is a primary measure in Moroccan agriculture, with various connections to human needs. Its estimation represents a troublesome task in light of its fundamental relation to crop market planning. Conventional techniques are simple, but require human effort and time. Advancement has been made in remote sensing by using deep learning architectures. This paper discusses different estimation techniques and presents a methodology to estimate wheat yield based on multimodal remote sensing data and exploiting 3D CNN (3-Dimensional Convolutional Neural Network) architecture which can be used to extract dynamic features over consecutive time. Unlike the 2D convolution kernel, which only moves in the two dimensions of height and width and passes through images horizontally and vertically, the 3D kernel establishes convolution while adding an additional dimension generally represented by time and therefore moves through these three dimensions. Our paper suggests this architecture as a methodological solution to develop a precise yield determination.