Agricultural monitoring system using images through a LPWAN network

Abstract

Internet of things (IoT) has turned into an opportunity to connect millions of devices through communication networks in digital environments. Inside IoT and mainly in the technologies of communication networks, it is possible to find Low Power Wide Area Networks (LPWAN). Within these technologies, there are service platforms in unlicensed frequency bands such as the LoRa Wide Area Network (LoRaWAN). It has features such as low power consumption, long-distance operation between gateway and node, and low data transport capacity. LPWAN networks are not commonly used to transport high data rates as in the case of agricultural images. The main goal of this research is to present a methodology to transport images through LPWAN networks using LoRa modulation. The methodology presented in this thesis is composed of three stages mainly.

The first one is image processing and classification process. This stage allows preparing the image in order to give the information to the classifier and separate the normal and abnormal images; i.e. to classify the images under the normal conditions of its representation in contrast with the images that can represent some sick or affectation with the consequent presence of a particular pathology. For this activity. it was used some techniques were used classifiers such as Support Vector Machine SVM, K-means clustering, neuronal networks, deep learning and convolutional neuronal networks. The last one offered the best results in classifying the samples of the images.

The second stage consists in a compression technique and reconstruction algorithms. In this stage, a method is developed to process the image and entails the reduction of the high amount of information that an image has in its normal features with the goal to transport the lowest amount of information. For this purpose, a technique will be presented for the representation of the information of an image in a common base that improves the reduction process of the information. For this activity, the evaluated components were Wavelet, DCT-2D and Kronecker algorithms. The best results were obtained by Wavelet Transform. On the other hand, the compres- sion process entails a series of iterations in the vector information, therefore, each iteration is a possibility to reduce that vector until a value with a minimum PSNR (peak signal to noise ratio) that allows rebuilding the original vector. In the reconstruction process, Iterative Hard Thresholding (IHT), Ortogonal MAtching Pur- suit (OMP), Gradient Projection for Sparse Reconstruction (GPSR)and Step Iterative Shrinage/Thresholding (Twist) algorithms were evaluated. Twist showed the best performance in the results.

Finally, in the third stage, LoRa modulation is implemented through the creation of LoRa symbols in Matlab with the compressed information. The symbols were delivered for transmission to Software Defined Radio (SDR). In the receptor, a SDR device receives the signal, which is converted into symbols that are in turn converted in an information vector. Then, the reconstruction process is carried out following the description in the last part of stage 2 - compression technique and reconstruction algorithms, which is described in more detailed in chapter 3, section 3.2. Finally, the image reconstructed is presented. The original image and the result image were compared in order to find the differences. This comparison used Peak Signal-to-Noise Ratio (PSNR) feature in order to get the fidelity of the reconstructed image with respect of the original image. In the receptor node, it is possible to observe the pathology of the leaf. The methodology is particularly applied for monitoring abnormal leaves samples in potato crops.

This work allows finding a methodology to communicate images through LPWAN using the LoRa modulation technique. In this work, a framework was used to classify the images, then, to process them in order to reduce the amount of data, to establish communication between a transmitter and a receiver through a wireless communication system and finally, in the receptor, to obtain a picture that shows the particularity of the pathology in an agricultural crop.