Preface

Artificial intelligence learning is the fastest-growing field in computer science. Deep learning algorithms and techniques are found to be useful in different areas, such as automatic machine translation, automatic handwriting generation, visual recognition, fraud detection, and detecting developmental delays in children. Deep Learning Techniques for Automation and Industrial Applications presents a concise introduction to the recent advances in the field of artificial intelligence (AI). The broad-ranging discussion herein covers the algorithms and applications in the areas of AI, reasoning, machine learning, neural networks, reinforcement learning, and their applications in various domains like agriculture and healthcare. Applying deep learning techniques or algorithms successfully in these areas requires a concerted effort, fostering integrative research between experts ranging from diverse disciplines, from data science to visualization.

This book provides state-of-the-art approaches to deep learning in these areas. It covers detection and prediction, as well as future framework development, building service systems, and analytical aspects. For all these topics, various approaches to deep learning, such as artificial neural networks, fuzzy logic, genetic algorithms, and hybrid mechanisms, are explained.

The successful application of deep learning techniques to enable meaningful, cost-effective, personalized cloud security service is a primary and current goal. However, realizing this goal requires effective understanding, application, and amalgamation of deep learning and several other computing technologies to deploy such a system effectively. This book helps to clarify certain key mechanisms of technology to realize a successful system. It enables the processing of very large datasets to help with precise and comprehensive forecasts of risk and delivers recommended actions that improve outcomes for consumers. This is a novel application domain of deep learning and is of prime importance to all of human civilization.

Preparing both undergraduate and graduate students for advanced modeling and simulation courses, this book helps them to carry out effective simulation studies. In addition, graduate students will be able to comprehend and conduct AI and data mining research after completing this book.

The book comprises fourteen chapters. In Chapter 1, images play a crucial role in describing, representing, and conveying information, which aids in human productivity, cost, analysis, and other areas. Text extraction is the method used to convert text to plain text. Text extraction is a very challenging problem because of the many changes in these texts’ size, orientation, and alignment, as well as low-resolution/pixelated images, noisy backgrounds, etc. Using the Tesseract OCR engine, we aim to reduce these issues in this project. Tesseract is developed by Google and its an opensource optical character recognition (OCR) engine. OCR technology allows computers to recognize text in images, making it possible to convert images of text into machine-readable text. Tesseract has been trained in a wide variety of languages and scripts, including English, Chinese, Arabic, and more.

Chapter 2 addresses agriculture, which is the most important part of everyone’s life. Crops and plants play a huge role in the making of life and taking care of those crops and plants is both important and tough. Thus, to detect the disease in plants, this chapter demonstrates how to tell whether a plant is healthy or not. The dataset consists of chilly plant leaves collected from one of the fields located in Andhra Pradesh, India. Many image classification models, as well as transfer learning models, are applied. Deep learning models CNN and transfer learning models, like InceptionV3 and VGG16, are applied with and without data augmentation.

In Chapter 3, researchers have applied various deep learning and transfer learning methods to accurately predict the disease of a damaged plant, so that we can cure the plant in its initial stage. The models are trained on the image dataset containing various categories of plants like mango and pomegranate. The results state that ResNet outperformed Inception, VGG19, and CNN by giving an accuracy of 88% and 87.5% percent for pomegranate and mango respectively.

In Chapter 4, researchers have compared different deep learning-based classification techniques on a remote sensing image dataset. The dataset has been taken from the UC Merced Land Use Dataset, which contains a total of 21 classes, with every class consisting of 100 images of size 256 x 256. The models used in this study are VGG, ResNet, Inception, Dense Net, and Efficient Net, which are deep convolutional network architectures for image classification with different numbers of layers. To make meaningful comparisons, all models were extended by adding three layers at the end to improve their performance. The performance of the VGG19 model was found to superior. This model was able to classify almost all images belonging to 21 classes with an accuracy of 100% in training and 95.07% in testing data, followed by VGG16 with 93% and ResNet with 91% accuracy in testing data.

Chapter 5 deals with sarcasm. Sarcasm is a sardonic or bitter remark, intended to express disrespect or ridicule. It is used in Hindi language, originating from many of Hindi idioms and proverbs, and often uses indirect sarcasm, as in saying, “” with the meaning, “.” Sentiment categorization is easier in comparison to sarcasm detection, as we see in the above-written idiom, which contains a negative sentiment. The intention of the sentence is to call an uneducated person knowledgeable among a group of fools. In the present scenario, people on social media platforms like Twitter, Facebook, and WhatsApp succeed in recognizing sarcasm despite interacting with strangers across the world. Sarcasm detection is a challenging task in Natural Language Processing due to the richness of morphology. Detecting sarcasm in Hindi language tweets is a prime task for Natural Language processing to avoid misconstruing sarcastic statements as original literal statements.

Chapter 6 explains how the image is the main source for all image processing fields, like surveillance, detection, recognition, satellite, etc. Good visibility of images captured by sensors becomes crucial for all computer vision tasks. Sometimes the scene quality is degraded by bad weather conditions like haze, fog, or smoke, therefore making it difficult for the computer vision area to obtain actual information. Haze can be removed from a single-input scene by using dehazing methods. Synthetic haze can be created by a haze generator, and, currently, most image dehazing techniques are applied for synthetic haze. Various single-image dehazing techniques are being developed and tested on real-world scenes that are captured in hazy environments using cameras.

Chapter 7 demonstrates how the framework needs accurate and realtime performance to count how many people are present at a particular moment in a particular frame. So, our counting framework automatically detects each person’s face and makes instantaneous decisions to count the number of persons in front of the camera or within a set of images. The work of individual counting can be done in two broad ways: the first is the detection of faces; the second is the counting approach used to track and count people within a frame.

Chapter 8 describes how CNN networks can show a resemblance to traditional steganalysis by using filters for feature extraction. Due to the use of content-adaptive steganographic methods, the stego message is hidden more often in the complex areas of the image and thus cannot be detected with a simple statistical analysis of the image. The stego information in these steganographic methods affect the dependencies between the pixels introduced through various kinds of noise present in the image. Thus, the difference between the cover and stego image is identified through the noise part rather than the image content. Different researchers have used various preprocessing filters for calculating the noise residuals and passing them to the CNN network instead of images directly. This work employs a content adaptive steganography method, Highly Undetectable Steganography (HUGO), for creating stego images. Furthermore, this work provides a comparative analysis of one of the variants of CNN models specific for steganalysis and various pre-trained models of computer vision that apply to steganalysis.

Chapter 9 discusses how groundwater abstraction beyond the safe limit is causing a rapid groundwater table depletion at the rate of 1–2 m/year in many districts. Uncontained and unplanned usage may affect food production by 20 percent. Due to the significant impact of this imperceptible resource on various aspects of life, the economy, the environment, and society, there is a pressing need to enhance the scientific comprehension, estimation, and administration of groundwater management. A scientific framework for the demarcation of its potential storage and recharge zonal maps, i.e., GWPSZ and GWRZ, can be instrumental in this regard for urban and rural water committees to objectively manage the resources at the regional level.

Chapter 10 explains the process of using pre-trained models to classify different types of fruit leaves accurately. We also discuss the advantages of transfer learning for industrial applications, including improved accuracy, reduced training time, and better utilization of resources. We...