Ghiordy Contreras

- Spectral imaging provides valuable additional information that enhances various imaging applications, including biomedical imaging, culture identification, and surveillance.
- These applications leverage the unique features of spectral scenes captured through technologies like Coded Aperture Snapshot Spectral Imagers (CASSI), which inherently apply compressive sensing principles. However, the effectiveness of these methods is often compromised due to the practical limitations of the sensing matrix that deviate from ideal conditions.
- This publication introduces a deep learning-based approach to correct real-compressed measurements and estimate ideal-corrected measurements.
- The correction process is derived from a matrix representation of compressed measurements, incorporating compressed spatial dimensions and the number of projections captured as shots.
- Model performance is evaluated using metrics such as peak signal-to-noise ratio and structural similarity index, based on the reconstructed data cube utilizing the gradient projection for sparse reconstruction algorithm.
- The results demonstrate that the deep learning-based method significantly enhances reconstruction quality in comparison to the ground truth images, particularly when noise levels are manageable.

- Developed a novel application leveraging Computer Vision and Machine Learning techniques to identify k clusters within datasets that exhibit overlapping issues.
- Addressed challenges in unsupervised data clustering, where distinguishing between groups can be complex.
- Computed pairwise distance calculations on the original data to generate a Distances Matrix, which serves as a comprehensive representation of the dataset.
- Utilized morphological operators to extract significant features from the Distances Matrix, facilitating the individual identification of groups within the dataset.
- Conducted matrix decomposition on the covariance matrix, calculating data elements for each cluster to project the data into a new linear space.
- Achieved correction of overlapping and separation distances among clusters without compromising information integrity.
- Demonstrated accurate identification of k clusters, effectively eliminating data overlap while retaining all relevant information.
- Employed clustering validation metrics, such as Silhouette and Precision, to rigorously assess and validate the effectiveness of the proposed methodology.

- Developed a thermodynamic modeling system utilizing computer science to analyze the incubation processes of domestic birds, revealing significantly high energy consumption compared to the energy utilized in these processes.
- Conducted a comprehensive data analysis of temperature and relative humidity variables within heating zones to determine the efficiency of energy supplied by the source.
- Simultaneously measured voltage and current alongside temperature and relative humidity to gain a holistic understanding of the energy dynamics.
- Employed artificial neural networks to analyze data collected from sensors, addressing the highly time-variant nature of the real processes and establishing fixed environmental conditions as needed.
- Achieved an air flow measurement of 3.4375 x 10−2 m³/J using an anemometer, correlating this with the electrical energy supplied by fans, which averaged 9.4818 W using ceramic resistances.
- Tested an adaptive controller where variables were calibrated using equations derived from the data analysis to enhance system performance.
- Recognized the economic challenges faced by Colombian farmers due to recent free trade agreements, and developed this system using open-source software and hardware to minimize costs associated with licensing and subscriptions from proprietary products.