Enrico Casella

Enrico Casella

Auto-generated profile, based on publicly available information

Enrico Casella

Department of Animal Science, The Pennsylvania State University, State College, PA 16802

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 April 2024 in IEEE Sensors Letters
Verified
Journal Article
Closed access
HDHarish Devaraj
Harish Devaraj
SSShaleeza Sohail
Shaleeza Sohail
MOMelanie Ooi
Melanie Ooi
Boyang LiBoyang Li
Boyang Li
NHNathaniel Hudson
Nathaniel Hudson
Omer RanaOmer Rana
Omer Rana
Precision horticulture is evolving due to scalable sensor deployment and machine learning (ML) integration. These advancements boost the operational efficiency of individual farms, balancing the benefits of analytics with autonomy requirements. However, given concerns that affect wide geographic regions (e.g., climate change), there is a need to apply models that span farms. Federated learning (FL) has emerged as a potential solution. FL enables decentralized ML across different farms without sharing private data. Traditional FL assumes simple two-tier network topologies and, thus, falls short of operating on more complex networks found in real-world agricultural scenarios. Networks vary across crops and farms and encompass various sensor data modes, extending across jurisdictions. New hierarchical FL (HFL) approaches are needed for more efficient and context-sensitive model sharing, accommodating regulations across multiple jurisdictions. We present the RuralAI architecture deployment for tomato crop monitoring, featuring sensor field units for soil, crop, and weather data collection. HFL with personalization is used to offer localized and adaptive insights. Model management, aggregation, and transfers are facilitated via a flexible approach, enabling seamless communication between local devices, edge nodes, and the cloud.

30 June 2023 in IEEE Access
Verified
Journal Article
Open access
ECEnrico Casella
Enrico Casella
MCMelissa C. Cantor
Melissa C. Cantor
MSMegan M. Woodrum Setser
Megan M. Woodrum Setser
SSSimone Silvestri
Simone Silvestri
JCJoao H Costa
Joao H Costa
Bovine Respiratory Disease (BRD) is an infection of the respiratory tract that is the leading reason for antimicrobial use in dairy calves and represents 22% of calf mortalities. The costs and effects of BRD can severely damage a farm’s economy, since raising dairy calves is one of the largest economic investments, and diagnosing BRD requires intensive and specialized labor that is hard to find. Precision technologies based on the Internet of Things (IoT), such as automatic feeders, scales, and accelerometers, can help detect behavioral changes before outward clinical signs of BRD. Such early detection enables early treatment, and thus faster recovery, with less long term effects. In this paper, we propose a framework for BRD diagnosis, its early detection, and identification of BRD persistency status using precision IoT technologies. We adopt a machine learning model paired with a cost-sensitive feature selection problem called Cost Optimization Worth (COW). COW maximizes prediction accuracy given a budget constraint. We show that COWis NP-Hard, and propose an efficient heuristic with polynomial complexity called Cost-Aware Learning Feature (CALF). We validate our methodology on a real dataset collected from 159 calves during the preweaning period. Results show that our approach outperforms a recent state-of-the-art solution. Numerically, we achieve an accuracy of 88% for labeling sick and healthy calves, 70% of sick calves are predicted 4 days prior to diagnosis, and 80% of persistency status calves are detected within the first five days of sickness.

1 September 2020 in Pervasive and Mobile Computing
Verified
Journal Article
Closed access
ECEnrico Casella
Enrico Casella
AKAtieh R. Khamesi
Atieh R. Khamesi
SSSimone Silvestri
Simone Silvestri

1 September 2025 in Journal of Dairy Science
Verified
Journal Article
Open access
GMGuilherme L. Menezes
Guilherme L. Menezes
Anderson A. AlvesAnderson A. Alves
Anderson A. Alves
ANAriana Negreiro
Ariana Negreiro
RFRafael E.P. Ferreira
Rafael E.P. Ferreira
SHShogo Higaki
Shogo Higaki
JDJoão Ricardo. R. Dórea
João Ricardo. R. Dórea
Computer vision systems offer identification solutions for animals with distinct coat patterns but are less effective for solid-colored herds. Another complication in real-world conditions is when new animals are introduced or removed from the herd, representing a challenge known as the open-set scenario. A promising alternative for identifying solid-colored animals is to use keypoints, similar to recognition systems that use parts of the human face and body. However, body growth can alter biometric features over time, potentially affecting model accuracy. Therefore, this study aims to (1) evaluate the capacity to automatically identify dairy cows using keypoints at specific anatomical landmarks (i.e., bony prominences) and determine the number of images necessary to train the model; (2) assess the effect of body biometric features perturbations (i.e., a controlled increase in biometric feature size simulating body growth variations over time in the testing set) on the identification accuracy; and (3) evaluate the capacity to automatically identify dairy cows using keypoints in open-set scenarios. A pose estimation model was trained to identify 10 keypoints per cow, located at anatomical landmarks. The Euclidean distance between these keypoints was used to generate biometric features for each cow. This trained pose estimation model was used to predict keypoints on a total of 118 videos containing 41 individual cows. Three strategies were applied using all 10 keypoints over the whole body (strategy 1) or 8 keypoints on the back of the cow (strategy 2), simulating occlusion on the neck and head, or 6 keypoints from the rump (strategy 3) to investigate the accuracy of identifying the cow when only the rump is visible. These strategies resulted in a total of 3,143; 6,650; and 11,499 predicted frames, respectively. For cow identification, a Siamese neural network was trained using the biometric features extracted from the keypoints. Three experiments were tested: (1) closed set scenarios and sample size for model training; (2) closed set with perturbation of 2%, 4%, 7%, and 9%; and (3) open set considering 20% of the herd as new cows. In the closed set, the best result was observed using strategy 1, showing an accuracy, precision, recall, and F1-score of 96.3%, 96.3%, 96.0%, and 95.9%, respectively. Training the model with 18 images per class (i.e., individuals) achieved over 80% accuracy at the frame level. After applying perturbation in the closed set, only the strategy with 2% perturbation showed satisfactory results. In open-set scenarios, the model achieved an accuracy, precision, recall, and F1-score of 72.5%, 100%, 72.5%, and 84.1%, respectively, in identifying the new cows. Considering both the open and closed set scenarios without perturbation, the model showed an accuracy, precision, recall, and F1-score of 75.4%, 79.5%, 76.7%, and 75.8%, respectively. These results suggest that strategy 1, using Euclidean distances between the 10 keypoints, can effectively identify individual animals of similar color.

12 November 2025 in Poultry Science
Verified
Journal Article
Open access
GCGiulio Calderone
Giulio Calderone
MMMireia Molins
Mireia Molins
PCPietro Catania
Pietro Catania
JBJohn W. Boney
John W. Boney
ECEnrico Casella
Enrico Casella
Monitoring animal health and welfare in large commercial poultry houses is an essential aspect of management and productivity, but it is a time-consuming and labor-intensive task. Additionally, high employee turnover increases training costs and compromises effective and efficient operations. To address these challenges, this study explores the combined use of an Unmanned Aerial Vehicle (UAV) and computer vision technology as an innovative, non-invasive approach to support continuous behavioral monitoring and staff workload. Using a DJI Neo UAV equipped with a color camera, overhead videos of 160 Nicholas Select turkey toms from age 5 d to 32 d were acquired. Video frames were sampled and manually annotated to create a dataset for a You Only Look Once (YOLO) computer vision model capable of detecting and classifying eight key behaviors: feeding, drinking, sitting, standing, perching, huddling, wing flapping, and dead. The annotated dataset of 2388 images was divided into 70 % training, 15 % validation and 15 % testing. Among the tested configurations, the best model was selected based on the Mean Average Precision 50-95 (mAP50-95) performance on the validation set, corresponding to version 11 of the large YOLO model (YOLOv11-l) with an image resolution of 1280 pixels per side. Performance metrics for this model on the independent testing set were: precision = 0.90, recall = 0.87, F1-score = 0.89, and Mean Average Precision 50 (mAP50) = 0.89 using confidence threshold of 0.20. These results suggest that UAV-integrated computer vision could be applied to accurately classify turkey behaviors in commercial environments, improving welfare monitoring and management efficiency.

29 July 2024 in ACM Transactions on Cyber-Physical Systems
Verified
Journal Article
Open access
ECEnrico Casella
Enrico Casella
SSSimone Silvestri
Simone Silvestri
DBD. A. Baker
D. A. Baker
SSajal K. Das
Sajal K. Das
Extreme outside temperatures resulting from heat waves, winter storms, and similar weather-related events trigger the Heating Ventilation and Air Conditioning (HVAC) systems, resulting in challenging, and potentially catastrophic, peak loads. As a consequence, such extreme outside temperatures put a strain on power grids and may thus lead to blackouts. To avoid the financial and personal repercussions of peak loads, demand response and power conservation represent promising solutions. Despite numerous efforts, it has been shown that the current state-of-the-art fails to consider (1) the complexity of human behavior when interacting with power conservation systems and (2) realistic home-level power dynamics. As a consequence, this leads to approaches that are (1) ineffective due to poor long-term user engagement and (2) too abstract to be used in real-world settings. In this article, we propose an auction theory-based power conservation framework for HVAC designed to address such individual human component through a three-fold approach: personalized preferences of power conservation, models of realistic user behavior , and realistic home-level power dynamics . In our framework, the System Operator sends Load Serving Entities (LSEs) the required power saving to tackle peak loads at the residential distribution feeder. Each LSE then prompts its users to provide bids , i.e., personalized preferences of thermostat temperature adjustments, along with corresponding financial compensations. We employ models of realistic user behavior by means of online surveys to gather user bids and evaluate user interaction with such system. Realistic home-level power dynamics are implemented by our machine learning-based Power Saving Predictions (PSP) algorithm, calculating the individual power savings in each user’s home resulting from such bids. A machine learning-based PSPs algorithm is executed by the users’ Smart Energy Management System (SEMS). PSP translates temperature adjustments into the corresponding power savings. Then, the SEMS sends bids back to the LSE, which selects the auction winners through an optimization problem called POwer Conservation Optimization (POCO). We prove that POCO is NP-hard, and thus provide two approaches to solve this problem. One approach is an optimal pseudo-polynomial algorithm called DYnamic programming Power Saving (DYPS), while the second is a heuristic polynomial time algorithm called Greedy Ranking AllocatioN (GRAN). EnergyPlus, the high-fidelity and gold-standard energy simulator funded by the U.S. Department of Energy, was used to validate our experiments, as well as to collect data to train PSP. We further evaluate the results of the auctions across several scenarios, showing that, as expected, DYPS finds the optimal solution, while GRAN outperforms recent state-of-the-art approaches.