Automated machine learning-based models for predicting and evaluating antibiotic removal in constructed wetlands.

Machine learning models can improve antibiotic removal performance in constructed wetlands (CWs) by optimizing the operation process. However, robust modeling approaches for revealing the complex biochemical treatment process of antibiotics in CWs are still lacking. In this study, two automated machine learning (AutoML) models achieved good performance with different sizes of the training dataset (mean absolute error = 9.94-13.68, coefficient of determination = 0.780-0.877), demonstrating the ability to predict antibiotic removal performance without human intervention. Explainable analysis results (the variable importance and Shapley additive explanations) revealed that the variable substrate type was more influential than the variables of influent wastewater quality and plant type. This study proposed a potential approach to comprehensively understanding the complex effects of key operational variables on antibiotic removal, which serve as a reference for optimizing operational adjustments in the CW process.

Molecular probing of dissolved organic matter and its transformation in a woolen textile wastewater treatment station.

Woolen textile industry produces enormous wastewater (WTIW) with high pollution loads, and needs to be treated by wastewater treatment stations (WWTS) before centralized treatment. However, WTIW effluent still contains many biorefractory and toxic substances; thus, comprehensive understandings of dissolved organic matter (DOM) of WTIW and its transformation are essential. In this study, total quantity indices, size exclusion chromatography, spectral methods, and Fourier transform ion cyclotron resonance mass spectrometry (FTICR MS) were used for comprehensively characterizing DOM and its transformation during full-scale treatments, including influent, regulation pool (RP), flotation pool (FP), up-flow anaerobic sludge bed (UA), anaerobic/oxic (AO) and effluent. DOM in influent featured a large molecular weight (5-17 kDa), toxicity (0.201 HgCl2 mg/L), and a protein content of 338 mg C/L. FP largely removed 5-17 kDa DOM with the formation of 0.45-5 kDa DOM. UA and AO removed 698 and 2042 chemicals, respectively, which were primarily saturated components (H/C > 1.5); however, both UA and AO contributed to the formation of 741 and 1378 stable chemicals, respectively. Good correlations were found among water quality indices and spectral/molecular indices. Our study reveals the molecular composition and transformation of WTIW DOM during treatments and encourages the optimization of the employed processes in WWTS.

Carbon source recovery from waste sludge reduces greenhouse gas emissions in a pilot-scale industrial wastewater treatment plant.

Carbon cycle regulation and greenhouse gas (GHG) emission abatement within wastewater treatment plants (WWTPs) can theoretically improve sustainability. Currently, however, large amounts of external carbon sources used for deep nitrogen removal and waste sludge disposal aggravate the carbon footprint of most WWTPs. In this pilot-scale study, considerable carbon was preliminarily recovered from primary sludge (PS) through short-term (five days) acidogenic fermentation and subsequently utilized on-site for denitrification in a wool processing industrial WWTP. The recovered sludge-derived carbon sources were excellent electron donors that could be used as additional carbon supplements for commercial glucose to enhance denitrification. Additionally, improvements in carbon and nitrogen flow further contributed to GHG emission abatement. Overall, a 9.1% reduction in sludge volatile solids was achieved from carbon recovery, which offset 57.4% of external carbon sources, and the indirect GHG emissions of the target industrial WWTP were reduced by 8.05%. This study demonstrates that optimizing the allocation of carbon mass flow within a WWTP has numerous benefits.

Real-Time Small Drones Detection Based on Pruned YOLOv4.

To address the threat of drones intruding into high-security areas, the real-time detection of drones is urgently required to protect these areas. There are two main difficulties in real-time detection of drones. One of them is that the drones move quickly, which leads to requiring faster detectors. Another problem is that small drones are difficult to detect. In this paper, firstly, we achieve high detection accuracy by evaluating three state-of-the-art object detection methods: RetinaNet, FCOS, YOLOv3 and YOLOv4. Then, to address the first problem, we prune the convolutional channel and shortcut layer of YOLOv4 to develop thinner and shallower models. Furthermore, to improve the accuracy of small drone detection, we implement a special augmentation for small object detection by copying and pasting small drones. Experimental results verify that compared to YOLOv4, our pruned-YOLOv4 model, with 0.8 channel prune rate and 24 layers prune, achieves 90.5% mAP and its processing speed is increased by 60.4%. Additionally, after small object augmentation, the precision and recall of the pruned-YOLOv4 almost increases by 22.8% and 12.7%, respectively. Experiment results verify that our pruned-YOLOv4 is an effective and accurate approach for drone detection.