Abstract
Nowadays, social networks are one of the biggest ways of sharing real time information. These networks, have several groups focused on sharing information about road incidents and other traffic events. The work here presented aims the creation of an AI model capable of identifying publications related to traffic events in a specific road, based on publications shared on social networks. A predictive model was obtained by training a deep learning model for the detection of publications related with road incidents with an average accuracy of 95%. The model deployed as a service is already fully functional and is operating in 24/7 while awaits a final integration with the road management system of a company where it will be used to support the Control Center team in the decision making. © 2022 Elsevier B.V.. All rights reserved.

Abstract
Object identification, such as tree trunk detection, is fundamental for forest robotics. Intelligent vision systems are of paramount importance in order to improve robotic perception, thus enhancing the autonomy of forest robots. To that purpose, this paper presents three contributions: an open dataset of 5325 annotated forest images; a tree trunk detection Edge AI benchmark between 13 deep learning models evaluated on four edge-devices (CPU, TPU, GPU and VPU); and a tree trunk mapping experiment using an OAK-D as a sensing device. The results showed that YOLOR was the most reliable trunk detector, achieving a maximum F1 score around 90% while maintaining high scores for different confidence levels; in terms of inference time, YOLOv4 Tiny was the fastest model, attaining 1.93 ms on the GPU. YOLOv7 Tiny presented the best trade-off between detection accuracy and speed, with average inference times under 4 ms on the GPU considering different input resolutions and at the same time achieving an F1 score similar to YOLOR. This work will enable the development of advanced artificial vision systems for robotics in forestry monitoring operations.

Abstract
Physical activity is movement of the body or part of the body to make muscles more active and to lose the energy from the body. Regular physical activity in the daily routine is very important to maintain good physical and mental health. It can be performed at home, a rehabilitation center, gym, etc., with a regular monitoring system. How long and which physical activity is essential for specific people is very important to know because it depends on age, sex, time, people that have specific diseases, etc. Therefore, it is essential to monitor physical activity either at a physical activity center or even at home. Physiological parameter monitoring using contact sensor technology has been practiced for a long time, however, it has a lot of limitations. In the last decades, a lot of inexpensive and accurate non-contact sensors became available on the market that can be used for vital sign monitoring. In this study, the existing research studies related to the non-contact and video-based technologies for various physiological parameters during exercise are reviewed. It covers mainly Heart Rate, Respiratory Rate, Heart Rate Variability, Blood Pressure, etc., using various technologies including PPG, Video analysis using deep learning, etc. This article covers all the technologies using non-contact methods to detect any of the physiological parameters and discusses how technology has been extended over the years. The paper presents some introductory parts of the corresponding topic and state of art review in that area.

Abstract
The development of foot ulcers is associated with the Diabetic Foot (DF), which is a problem detected in patientswith Diabetes Mellitus (DM). Several studies demonstrate that thermography is a technique that can be used to identify and monitor the DF problems, thus helping to analyze the possibility of ulcers arising, as tissue inflammation causes temperature variation. There is great interest in developing methods to detect abnormal plantar temperature changes, since healthy individuals generally show characteristic patterns of plantar temperature variation and that the plantar temperature distribution of DF tissues does not followa specific pattern, so temperature variations are difficult to measure. In this sequel, a methodology, that uses thermograms to analyze the diversity of thermal changes that exist in the plant of a foot and classifies it as being from an individual with possibility of ulcer arising or not, is presented in this paper. Therefore, the concept of clustering is used to propose binary classifiers with different descriptors, obtained using two clustering algorithms, to predict the risk of ulceration in a foot. Moreover, for each descriptor, a numerical indicator and a classification thresholder are presented. In addition, using a combination of two different descriptors, a hybrid quantitative indicator is presented. A public dataset (containing 90 thermograms of the sole of the foot healthy people and 244 of DM patients) was used to evaluate the performance of the classifiers; using the hybrid quantitative indicator and the k-means clustering, the following metrics were obtained: Accuracy = 80%, AUC = 87% and F-measure = 86%.

Abstract
Melanoma is considered the deadliest type of skin cancer and in the last decade, the incidence rate has increased substantially. However, automatic melanoma classification has been widely used to aid the detection of lesions as well as prevent eventual death. Therefore, in this paper we decided to investigate how an attention mechanism combined with a classical backbone network would affect the classification of melanomas. This mechanism is known as triplet attention, a lightweight method that allows to capture cross-domain interactions. This characteristic helps to acquire rich discriminative feature representations. The different experiments demonstrate the effectiveness of the model in five different datasets. The model was evaluated based on sensitivity, specificity, accuracy, and F1-Score. Even though it is a simple method, this attention mechanism shows that its application could be beneficial in classification tasks.

Abstract
X-ray bone semantic segmentation is one crucial task in medical imaging. Due to deep learning's emergence, it was possible to build high-precision models. However, these models require a large quantity of annotated data. Furthermore, semantic segmentation requires pixel-wise labeling, thus being a highly time-consuming task. In the case of hip joints, there is still a need for increased anatomic knowledge due to the intrinsic nature of the femur and acetabulum. Active learning aims to maximize the model's performance with the least possible amount of data. In this work, we propose and compare the use of different queries, including uncertainty and diversity-based queries. Our results show that the proposed methods permit state-of-the-art performance using only 81.02% of the data, with O(1) time complexity.