Jiangsan Zhao
Research Scientist
Abstract
Raman spectroscopy provides detailed information about the molecular composition of a sample. The classical identification of components in a multi-component sample typically involves comparing the preprocessed spectrum with a known reference stored in a database using various spectral matching or machine-learning techniques or relies on universal models based on a two-step analysis including first, the component identification, and then the decomposition of the mixed signal. However, although large databases and universal models cover a wide range of target materials, they may be not optimized to the variability required in a specific application. In this study, we propose a single-step method using deep learning (DL) modeling to decompose a simulated mixture of real measurements of Raman scattering into relevant individual components regardless of noise, baseline and the number of components involved and quantify their ratios. We hypothesize that training a custom DL model for applications with a fixed set of expected components may yield better results than applying a universal quantification model. To test this hypothesis, we simulated 12,000 Raman spectra by assigning random ratios to each component spectrum within a library containing 13 measured spectra of organic solvent samples. One of the DL methods, a fully connected network (FCN), was designed to work on the raw spectra directly and output the contribution of each component of the library to the input spectrum in form of a component ratio. The developed model was evaluated on 3600 testing spectra, which were simulated similarly to the training dataset. The average component identification accuracy of the FCN was 99.7%, which was significantly higher than that of the universal custom trained DeepRaman model, which was 83.1%. The average mean absolute error for component ratio quantification was 0.000562, over one order of magnitude smaller than that of a well-established non-negative elastic net (NN-EN), which was 0.00677. The predicted non-zero ratio values were further used for component identification. Under the assumption that the components of a mixture are from a fixed library, the proposed method preprocesses and decomposes the raw data in a single step, quantifying every component in a multicomponent mixture, accurately. Notably, the single-step FCN approach has not been implemented in the previously reported DL studies.
Abstract
Fish counting is crucial in fish farming. Density map-based fish counting methods hold promise for fish counting in high-density scenarios; however, they suffer from ineffective ground truth density map generation. High labeling complexities and disturbance to fish growth during data collection are also challenging to mitigate. To address these issues, LDNet, a versatile network with attention implemented is introduced in this study. An imbalanced Optimal Transport (OT)-based loss function was used to effectively supervise density map generation. Additionally, an Image Manipulation-Based Data Augmentation (IMBDA) strategy was applied to simulate training data from diverse scenarios in fixed viewpoints in order to build a model that is robust to different environmental changes. Leveraging a limited number of training samples, our approach achieved notable performances with an 8.27 MAE, 9.97 RMSE, and 99.01% Accuracy on our self-curated Fish Count-824 dataset. Impressively, our method also demonstrated superior counting performances on both vehicle count datasets CARPK and PURPK+, and Penaeus_1k Penaeus Larvae dataset when only 5%–10% of the training data was used. These outcomes compellingly showcased our proposed approach with a wide applicability potential across various cases. This innovative approach can potentially contribute to aquaculture management and ecological preservation through counting fish accurately.
Abstract
Accurate locating and counting of litopenaeus vannamei fry can provide substantial support for vannamei fry sales and scientific feeding. However, traditional methods not only require visual observation by experts, but also are time-consuming and labor-intensive, with no guarantee to reach consensus between salesmen and customers. In contrast, more innovative methods require more expensive equipment or are only effective under specific conditions. The small size and high density nature of the shrimp fry makes its counting even more challenging. In this study, a point prediction method for counting and localization of litopenaeus vannamei fry with region-based super-resolution enhancement (PPCL-RSE) is proposed. Through the inclusion of three modules of density partitioning, high-density region expansion and regional super-resolution, the accuracy of fry counting and locating is improved. The model is deployed on a cloud server for convenient fry counting and localization based on images taken by smartphone cameras. To achieve this, we create a dataset called Vannamei-983 which contains images with more than 1,000,000 fry labeled. The proposed method shows accuracies of 99.04 % and 97.71 % in counting and localization of shrimp fry in low- and high-density images, respectively. The excellent model performance also demonstrate the effectiveness of the strategies considered in the study.
