Artificial intelligence based photonics techniques for agricultural applications

Abstract

Agriculture plays a critical role in the economic growth of every country and it is also considered as the backbone of economic system for developing nations. Agriculture is the largest source of livelihood in India and fulfills the basic human need. However, there are various factors that affect the crop yield and deteriorate the quality of production. Some of the most common factors that drastically affect the crop yield include quality of seeds, seed germination properties, viability, vigor, and impact of moisture content, salinity of soil, seed aging, temperature, and diseases. Traditional methods (germination test, field analysis, vigor test, chemical test) used to evaluate these properties and impact of various parameters on crop production are criticized as being tedious, manual, time consuming, inaccurate, destructive, laborious and requiring trained manpower. Moreover, image processing techniques utilizes 2D visual feature and do not characterize any biological parameter related to the properties of seeds and plants. Furthermore, spectral techniques also possess several disadvantages such as high cost, high computational complexity, requirement of sensitive detectors, and large data storage capabilities to acquire and process the data. Moreover, all these methods are limited to some specific application and cannot provide an optimum solution to cater all the aspect of crop managements and monitoring. In this direction, scope of photonics techniques including laser backscattering and biospeckle is explored for various agricultural applications. Both these techniques possess several advantages such as automatic operation, non-destructive and non-invasive testing, high speed, cost effectiveness, requirement of lesser number of components for experimental setup, simple experimental procedures, and its ability to get commercialized. Contemporary, image processing techniques used for analysis of data obtained by both the techniques possess several drawbacks and are dependent on various experimental and analysis procedures. Therefore, to process the data acquired from the various experiments, artificial intelligence (AI) based approach is developed to enable accurate understanding of the various factors that affects the crop yield. In this direction, efficient processing pipeline is developed for analysis of optical data by using various machine learning (ML) (support vector machine (SVM), logistic regression (LR), decision tree (DT), K-Nearest Neighbor (KNN), and Na¨ıve Bayes (NB)); and deep learning (DL) (artificial neural network (ANN), convolution neural network (CNN), 3D CNN, transfer learning (TL) models (VGG16, VGG19, ResNet50, and InceptionV3), and recurrent neural network (RNN) (long short-term memory (LSTM)) based algorithms. In this thesis, AI based photonics techniques is proposed for characterization of multiple factors and their impact on crop yield including pre-treatments analysis, seed quality identification, early identification of disease in seeds and plants for crop monitoring and protection. Firstly, laser biospeckle technique for characterization of priming treatments is proposed. Priming is one of the well-established and low-cost pre-sowing treatments for improving seed germination properties and productivity. The effect of different priming treatments on seed imbibition behaviour is evaluated and the results are benchmarked with the standard germination test. The biospeckle activity (BA) is found to be strongly correlated with the germination percentage (R=0.88, p<0.01) and negatively correlated with the mean germination time (R=-0.92, p<0.01).