Image Classification

● In Material Science, the ability to analyze microstructures relies on finding the reliable description and classification of the structures, which identify the microstructural image in order to determine the influence on mechanical properties, such as the maximum load that a body can support before breaking out. In almost all real solutions, microstructures are characterized by human experts, and its automatic identification is still a challenge. In order to demonstrate how the artificial neural network depth influences the accuracy achieved in microstructural classification, this work explores and compares four outstanding and standard Convolutional Neuronal Networks architectures.●In agriculture, specially, in Ecuador there are not automatic methods to provide an early detection against diseases or pests in crops. Based on this gap, the proposed work aims to identify leaf diseases from pixel images by using deep learning techniques.

Segmentation on computed tomography images

Segmentation techniques are focused on extracting the shapes of the object of interest with their respective borderlines (foreground pixels) from the background pixels.●In the recent years, soft tissues segmentation, such as colon segmentation has increased its applications in modern medicine fields, which use Computer Aided Diagnosis System (CAD) in order to reduce the dependence of diagnosis by doctors’ knowledge and experience to locate the prior tissue lesions timely and effectively. In this way, the colon segmentation in human abdominal CT images is the basis of analysis and identification of cancer nidus, providing powerful information in a CAD, such as early polyp detection, which can reduce the incidence of colon cancer. Moreover, the colon segmentation also may be introduced to make preoperative planning and simulations of general surgery. Therefore, a new approach for colon tissues segmentation on CT images which takes advantages of using deep and hierarchical learning about colon features with convolutional neural networks (CNN) is proposed.

Object Detection for embedded systems

● Object detection is based on determining if an object belonging to a specific class (object of interest as vehicles, animals, people, abandoned objects and many more) is present in the image or video sequences, and where in the image this object is located. Object detection consists of two main tasks: classification and localization. In this sense, the work is focused on implementing YOLO (You only look once) method within Raspberry Pi. YOLO convolutional neural network makes predictions with only one network evaluation. Its fast prediction and low computational complexity due to their low depth will be exploited in the Raspberry Pi board as an efficient low cost object detection system able to acquire and process video sequences without requiring external processing units.

Video scene classification

One of the challenges in AI is the ability of understanding video sequences to automatically detect and recognize patterns as well as classify video scenes (frame sequences that represent a real-world view containing multiples surfaces and objects) from pixel analysis. In this sense, over the past few years, deep learning techniques based on Deep Neural Networks (DNNs) have been introduced to face with the automatic scene understanding challenge. ● This proposed research is focused on introducing a new DNN architecture able to learn and identify video scenes from surveillance cameras data, after exploring and comparing the most outstanding CNNs architectures to analyze and take their main characteristics that allow to understand and classify correctly video sequence scenes. Besides the capacity of identification, the proposed approach will also be able to show an alert on the screen with information about the scene happening in order to provide a tool to ensure the safety of Ecuadorian citizens. This approach will work as the basis to create a new service to be implemented alongside with the surveillance cameras from the Ecu911 Security service (Integrated Security Service of Ecuador).