Machine Learning Based Crop Yield Prediction Using Spectral Images

Abstract

The socio-economic stability of a country heavily dependent on its agricultural outputs. Therefore, each country needs to monitor and maintain agricultural outcomes at an adequate level. The early prediction of crop yield helps the farmers adopt necessary changes in cultivation on a season and ensure food security. The crop yield depends on several parameters, such as vegetation parameters, climatic parameters, soil condition, etc. Spatial and temporal analysis of cropland is necessary for the accurate prediction of yield. The data for such analysis were collected with the help of regular field surveys. Such surveys required more human resources and lack accuracy due to the interpolation method adopted to map the readings to a larger geographical area. The advancement in satellite imaging techniques helps gather temporal data of broad geographical regions with less workforce. Usage of multispectral sensors in remote sensing helped in accurate discrimination of land objects and vegetations. The higher number of contiguous bands in hyperspec- tral images(HSI) improve the reconstruction of spectral signature and thereby increase the discrimination power. However, the higher dimensionality nature of HSI increases the computational complexity and leads to the Hughes phenomenon. The evolution of deep learning techniques made a significant impact on HSI classification. Several HSI processing applications rely on various Convolutional Neural Network (CNN) models. Therefore most of the CNN models perform dimensionality reduction (DR) as a pre- processing step. Another challenge in HSI classification is the consideration of both spatial and spectral features for obtaining accurate results. A few 3-D-CNN models are designed to overcome this challenge, but it takes more execution time than other methods. This research work proposes a multiscale spatio-spectral feature-based hy- brid CNN model for hyperspectral image classification. Hybrid DR used for optimal band extraction, which performs linear Gaussian Random Projection (GRP) and non- linear Kernel Principal Component Analysis (KPCA). A novel crop yield prediction model for the Paddy from Moderate Resolution Imaging Spectroradiometer (MODIS) data and climatic parameters is introduced in this research work. Various vegetation in- dices (VI) are collected from MODIS data for the crop’s entire life cycle. The proposed Temporal Convolutional network (TCN) with a specially designed dilated convolution module predicts the rice crop yield from vegetation indices and climatic parameters. The causal property of TCN and dilated convolution contribute to the multivariate time- based analysis of the crop and results in better performance.