Convolutional Network Research Articles

An Efficient 3D Convolutional Neural Network for Dose Prediction in Cancer Radiotherapy from CT Images

Introduction: Cancer is a highly lethal disease with a significantly high mortality rate. One of the most commonly used methods for treatment is radiation therapy. However, cancer treatment using radiotherapy is a time-consuming process that requires significant manual work from planners and doctors. In radiation therapy treatment planning, determining the dose distribution for each of the regions of the patient’s body is one of the most difficult and important tasks. Nowadays, artificial intelligence has shown promising results in improving the quality of disease treatment, particularly in cancer radiation therapy. Objectives: The main objective of this study is to build a high-performance deep learning model for predicting radiation therapy doses for cancer and to develop software to easily manipulate and use this model. Materials and Methods: In this paper, we propose a custom 3D convolutional neural network model with a U-Net-based architecture to automatically predict radiation doses during cancer radiation therapy from CT images. To ensure that the predicted doses do not have negative values, which are not valid for radiation doses, a rectified linear unit (ReLU) function is applied to the output to convert negative values to zero. Additionally, a proposed loss function based on a dose–volume histogram is used to train the model, ensuring that the predicted dose concentrations are highly meaningful in terms of radiation therapy. The model is developed using the OpenKBP challenge dataset, which consists of 200, 100, and 40 head and neck cancer patients for training, testing, and validation, respectively. Before the training phase, preprocessing and augmentation techniques, such as standardization, translation, and flipping, are applied to the training set. During the training phase, a cosine annealing scheduler is applied to update the learning rate. Results and Conclusions: Our model achieved strong performance, with a good DVH score (1.444 Gy) on the test dataset, compared to previous studies and state-of-the-art models. In addition, we developed software to display the dose maps predicted by the proposed model for each 2D slice in order to facilitate usage and observation. These results may help doctors in treating cancer with radiation therapy in terms of both time and effectiveness.

LRNet: Link Residual Neural Network for Blood Vessel Segmentation in OCTA Images.

Optical coherence tomography angiography (OCTA) is an emerging, non-invasive technique increasingly utilized for retinal vasculature imaging. Analysis of OCTA images can effectively diagnose retinal diseases, unfortunately, complex vascular structures within OCTA images possess significant challenges for automated segmentation. A novel, fully convolutional dense connected residual network is proposed to effectively segment the vascular regions within OCTA images. Firstly, a dual-branch structure Recurrent Residual Convolutional Neural Network (RRCNN) block is constructed utilizing RecurrentBlock and convolutional operations. Subsequently, the ResConvNeXt V2 Block is built as the backbone structure of the network. The output from the ResConvNeXt V2 Block is then fed into the side branch and the next ResConvNeXt V2 Block. Within the side branch, the Group Receptive Field Block (GRFB) processes the results from the previous and current layers. Ultimately, the side branch results are added to the backbone network outputs to produce the final segmentation. The model achieves superior performance. Experiments were conducted on the ROSSA and OCTA-500 datasets, yielding Dice scores of 91.88%, 91.72%, and 89.18% for the respective datasets, and accuracies of 98.31%, 99.02%, and 98.02%.

Deep Learning for Identification and Characterization of Ca ii Absorption Lines: A Multitask Convolutional Neural Network Approach

Quasar absorption lines are a powerful tool for studying the Universe, enabling us to probe distant gas, dust, and galaxy formation and evolution. However, detecting these lines, particularly Ca ii absorption lines, is a time-consuming and laborious process. Existing deep learning methods are prone to false positives and still require extensive manual verification and parameter measurement. This work presents three multitask convolutional neural network models and identifies the ResNet-CBAM model, which incorporates residual learning and an attention mechanism as the most effective. The results show that the ResNet-CBAM model achieves an accuracy of 99.7% in detecting Ca ii absorbers and excels in predicting critical parameters such as equivalent width and full width at half-maximum, with average correlation coefficients of 0.98 and 0.85, respectively. Furthermore, its remarkable generalization ability significantly improves detection precision on unseen data, rising from 20.3% of the cutting-edge model to 92.6%. In addition, with our numerous optimizations, our method can directly search for nonnormalized data, still achieving an accuracy of 98.6%. This translates to a dramatic reduction in manual inspection workload, paving the way for efficient and automated Ca ii absorber identification. In real-world applications on the Sloan Digital Sky Survey DR7 and DR12, our model successfully rediscovered 321 known Ca ii absorbers while identifying potential candidates in an additional 381 spectra. The codes used in this paper are available on Zenodo at doi:10.5281/zenodo.13953656.

Review and Intercomparison of Machine Learning Applications for Short-term Flood Forecasting

Abstract Among natural hazards, floods pose the greatest threat to lives and livelihoods. To reduce flood impacts, short-term flood forecasting can contribute to early warnings that provide communities with time to react. This manuscript explores how machine learning (ML) can support short-term flood forecasting. Using two methods [strengths, weaknesses, opportunities, and threats (SWOT) and comparative performance analysis] for different forecast lead times (1–6, 6–12, 12–24, and 24–48 h), we evaluate the performance of machine learning models in 94 journal papers from 2001 to 2023. SWOT reveals that the best short-term flood forecasting was produced by hybrid, random forest (RF), long short-term memory (LSTM), artificial neural network (ANN), and adaptive neuro-fuzzy inference system (ANFIS) approaches. The comparative performance analysis, meanwhile, favors convolutional neural network, ANFIS, multilayer perceptron, k-nearest neighbors algorithm (KNN), hybrid, LSTM, ANN, and support vector machine (SVM) at 1–6 h; hybrid, ANFIS, ANN, and LSTM at 6–12 h; SVM, hybrid, and RF at 12–24 h; and hybrid and RF at 24–48 h. In general, hybrid approaches consistently perform well across all lead times. Trends such as hybridization, model selection, input data selection, and decomposition seem to improve the accuracy of models. Furthermore, effective stand-alone ML models such as ANN, SVM, RF, genetic algorithm, KNN, and LSTM, provide better outcomes through hybridization with other ML models. By including different machine learning models and parameters such as environmental, socio-economical, and climatic parameters, the hybrid system can produce more accurate flood forecasting, making it more effective for early warning operational purposes.

Emotion Recognition from EEG Signals Using Advanced Transformations and Deep Learning

Affective computing aims to develop systems capable of effectively interacting with people through emotion recognition. Neuroscience and psychology have established models that classify universal human emotions, providing a foundational framework for developing emotion recognition systems. Brain activity related to emotional states can be captured through electroencephalography (EEG), enabling the creation of models that classify emotions even in uncontrolled environments. In this study, we propose an emotion recognition model based on EEG signals using deep learning techniques on a proprietary database. To improve the separability of emotions, we explored various data transformation techniques, including Fourier Neural Networks and quantum rotations. The convolutional neural network model, combined with quantum rotations, achieved a 95% accuracy in emotion classification, particularly in distinguishing sad emotions. The integration of these transformations can further enhance overall emotion recognition performance.

Sound recurrence analysis for acoustic scene classification

In everyday life, people experience different soundscapes in which natural sounds, animal noises, and man-made sounds blend together. Although there have been several studies on the importance of recurring sound patterns in music and language, the relevance of this phenomenon in natural soundscapes is still largely unexplored. In this article, we study the repetition patterns of harmonic and transient sound events as potential cues for acoustic scene classification (ASC). In the first part of our study, our aim is to identify acoustic scene classes that exhibit characteristic sound repetition patterns concerning harmonic and transient sounds. We propose three metrics to measure the overall prevalence of sound repetitions as well as their repetition periods and temporal stability. In the second part, we evaluate three strategies to incorporate self-similarity matrices as an additional input feature to a convolutional neural network architecture for ASC. We observe the characteristic repetition of transient sounds in recordings of “park” and “street traffic” as well as harmonic sound repetitions in acoustic scene classes related to public transportation. In the ASC experiments, hybrid network architectures, which combine spectrogram features and features from sound recurrence analysis, show increased accuracy for those classes with prominent sound repetition patterns. Our findings provide additional perspective on the distinctions among acoustic scenes previously primarily ascribed in the literature to their spectral features.

Enhanced Prediction of CO2-Brine Interfacial Tension at Varying Temperature Using a Multibranch-Structure-Based Neural Network Approach.

Interfacial tension (IFTC-B) between CO2 and brine depends on chemical components in multiphase systems, intricately evolving with a change in temperature. In this study, we developed a convolutional neural network with a multibranch structure (MBCNN), which, in combination with a compiled data set containing measurement data of 1716 samples from 13 available literature sources at wide temperature and pressure ranges (273.15-473.15 K and 0-70 MPa), was used to quantitatively explore the correlation of various chemical components with IFTC-B at varying temperature, aiming to achieve accurate predictions of IFTC-B under complex conditions. Our multibranch neural network analysis yielded some important insights: (1) Leveraging the convolutional and multibranch structure, MBCNN effectively mitigates the adverse effects of sparse matrices resulting from the absence of certain basic components, exhibiting higher prediction accuracy particularly for low IFTC-B scenarios (MAE = 0.47, and R2 = 0.9921) than other AI models. (2) The multibranch structure allows MBCNN to additionally capture the interattribute relationship between temperature and each chemical component. Such interattribute relationships are quantitatively correlated with IFTC-B, demonstrating that varying temperature significantly influences the dependence of IFTC-B on chemical components in gas and brine by causing the variation in their solubility. Specifically, the ratio of IFTC-B to the molality of monovalent cations (Na+ and K+) and bivalent cations (Ca2+ and Mg2+) in brine, as well as to the mole fraction of non-CO2 components (CH4 and N2) in the gas phase, varies with increasing temperature, approximately following a quadratic function. (3) By formulating the effect of each attribute on IFTC-B and quantifying their respective weight, we derived a new piecewise function for predicting IFTC-B at three temperature intervals (T ≤ 293.15 K, 293.15 K < T ≤ 324.4 K, and T > 324.4 K), with high prediction performance (MAE = 2.3672, R2 = 0.9263) across a wide temperature range in saline aquifers.

RL-MUL 2.0: Multiplier Design Optimization with Parallel Deep Reinforcement Learning and Space Reduction

Multiplication is a fundamental operation in many applications, and multipliers are widely adopted in various circuits. However, optimizing multipliers is challenging due to the extensive design space. In this paper, we propose a multiplier design optimization framework based on reinforcement learning. We utilize matrix and tensor representations for the compressor tree of a multiplier, enabling seamless integration of convolutional neural networks as the agent network. The agent optimizes the multiplier structure using a Pareto-driven reward customized to balance area and delay. Furthermore, we enhance the original framework with parallel reinforcement learning and design space pruning techniques and extend its capability to optimize fused multiply-accumulate (MAC) designs. Experiments conducted on different bit widths of multipliers demonstrate that multipliers produced by our approach outperform all baseline designs in terms of area, power, and delay. The performance gain is further validated by comparing the area, power, and delay of processing element arrays using multipliers from our approach and baseline approaches.

A Systematic Method Combining Rotated Convolution and State Space Augmented Transformer for Digitizing and Classifying Paper ECGs

Billions of paper Electrocardiograms (ECGs) are recorded annually worldwide, particularly in the Global South. Manual review of this massive dataset is time-consuming and inefficient. Accurate digital reconstruction of these records is essential for efficient cardiac disease diagnosis. This paper proposes a systematic framework for digitizing paper ECGs with 12 symmetrically distributed leads and identifying abnormal samples. This method consists of three main components. First, we introduce an adaptive rotated convolution network to detect the positions of lead waveforms. By exploiting the symmetric distribution of 12 leads, a novel loss is proposed to improve the detection model’s performance. Second, image processing techniques, including denoising and connected component analysis, are employed to digitize ECG waveforms. Finally, we propose a transformer-based classification method combined with a state space model. Our process is evaluated on a large synthetic dataset, including ECG images characterized by rotations, noise, and creases. The results demonstrate that the proposed detection method can effectively reconstruct paper ECGs, achieving an 11% improvement in SNR compared to the baseline. Moreover, our classification model exhibits slightly higher performance than other counterparts. The proposed approach offers a promising solution for the automated analysis of paper ECGs, supporting clinical decision-making.

A Data-Driven Deep Learning Framework for Prediction of Traffic Crashes at Road Intersections

Traffic crash prediction (TCP) is a fundamental problem for intelligent transportation systems in smart cities. Improving the accuracy of traffic crash prediction is important for road safety and effective traffic management. Owing to recent advances in artificial neural networks, several new deep-learning models have been proposed for TCP. However, these works mainly focus on accidents in regions, which are typically pre-determined using a grid map. We argue that TCP for roads, especially for crashes at or near road intersections which account for more than 50% of the fatal or injury crashes based on the Federal Highway Administration, has a significant practical and research value and thus deserves more research. In this paper, we formulate TCP at Road Intersections as a classification problem and propose a three-phase data-driven deep learning model, called Road Intersection Traffic Crash Prediction (RoadInTCP), to predict traffic crashes at intersections by exploiting publicly available heterogeneous big data. In Phase I we extract discriminative latent features called topological-relational features (tr-features), of intersections using a neural network model by exploiting topological information of the road network and various relationships amongst nearby intersections. In Phase II, in addition to tr-features which capture some inherent properties of the road network, we also explore additional thematic information in terms of environmental, traffic, weather, risk, and calendar features associated with intersections. In order to incorporate the potential correlation in nearby intersections, we utilize a Graph Convolution Network (GCN) to aggregate features from neighboring intersections based on a message-passing paradigm for TCP. While Phase II serves well as a TCP model, we further explore the signals embedded in the sequential feature changes over time for TCP in Phase III, by exploring RNN or 1DCNN which have known success on sequential data. Additionally, to address the serious issues of imbalanced classes in TCP and large-scale heterogeneous big data, we propose an effective data sampling approach in data preparation to facilitate model training. We evaluate the proposed RoadInTCP model via extensive experiments on a real-world New York City traffic dataset. The experimental results show that the proposed RoadInTCP robustly outperforms existing methods.

Expert Comment Generation Considering Sports Skill Level Using a Large Multimodal Model with Video and Spatial-Temporal Motion Features

In sports training, personalized skill assessment and feedback are crucial for athletes to master complex movements and improve performance. However, existing research on skill transfer predominantly focuses on skill evaluation through video analysis, addressing only a single facet of the multifaceted process required for skill acquisition. Furthermore, in the limited studies that generate expert comments, the learner’s skill level is predetermined, and the spatial-temporal information of human movement is often overlooked. To address this issue, we propose a novel approach to generate skill-level-aware expert comments by leveraging a Large Multimodal Model (LMM) and spatial-temporal motion features. Our method employs a Spatial-Temporal Attention Graph Convolutional Network (STA-GCN) to extract motion features that encapsulate the spatial-temporal dynamics of human movement. The STA-GCN classifies skill levels based on these motion features. The classified skill levels, along with the extracted motion features (intermediate features from the STA-GCN) and the original sports video, are then fed into the LMM. This integration enables the generation of detailed, context-specific expert comments that offer actionable insights for performance improvement. Our contributions are twofold: (1) We incorporate skill level classification results as inputs to the LMM, ensuring that feedback is appropriately tailored to the learner’s skill level; and (2) We integrate motion features that capture spatial-temporal information into the LMM, enhancing its ability to generate feedback based on the learner’s specific actions. Experimental results demonstrate that the proposed method effectively generates expert comments, overcoming the limitations of existing methods and offering valuable guidance for athletes across various skill levels.

Establishing a GRU-GCN coordination-based prediction model for miRNA-disease associations

BackgroundmiRNAs (microRNAs) are endogenous RNAs with lengths of 18 to 24 nucleotides and play critical roles in gene regulation and disease progression. Although traditional wet-lab experiments provide direct evidence for miRNA-disease associations, they are often time-consuming and complicated to analyze by current bioinformatics tools. In recent years, machine learning (ML) and deep learning (DL) techniques are powerful tools to analyze large-scale biological data. Hence, developing a model to predict, identify, and rank connections in miRNAs and diseases can significantly enhance the precision and efficiency in investigating the relationships between miRNAs and diseases.ResultsIn this study, we utilized miRNA-disease association data obtained by biotechnological experiments to develop a DL model for miRNA-disease associations. To improve the accuracy of prediction in this model, we introduced two labeling strategies, weight-based and majority-based definitions, to classify miRNA-disease associations. After preprocessing, data was trained with a novel model combining gated recurrent units (GRU) and graph convolutional network (GCN) to predict the level of miRNA-disease associations. The miRNA-disease association datasets were from HMDD (the Human miRNA Disease Database) and categorized by two distinct labeling approaches, weight-based definitions and majority-based definitions. We classified the miRNA-disease associations into three groups, “upregulated”, “downregulated” and “nonspecific”, by regression analysis and multiclass classification. This GRU-GCN coordinated model achieved a robust area under the curve (AUC) score of 0.8 in all datasets, demonstrating the efficacy in predicting potential miRNA-disease relationships.ConclusionsBy introducing innovative label-preprocessing methods, this study addressed the relationships between miRNAs and diseases, and improved the ambiguity of the results in different experiments. Based on these refined label definitions, we developed a DL-based model to refine and predict the results of associations between miRNAs and diseases. This model offers a valuable tool for complementing traditional experimental methods and enhancing our understanding of miRNA-related disease mechanisms.

Pneumothorax Detection System in Thoracic Radiography Images Using CNN Method

Purpose: This research aims to develop an automatic pneumothorax detection system using Convolutional Neural Networks (CNN) to classify thoracic radiography images. By leveraging CNN's effectiveness in identifying medical abnormalities, the system seeks to enhance diagnostic accuracy, reduce evaluation time, and minimize subjective interpretation errors. The output will provide a predicted label of "pneumothorax" or "non-pneumothorax," facilitating faster clinical treatment and improving diagnostic services while supporting radiologists in making more accurate and efficient decisions for this critical condition. Methods: This research employs an experimental deep learning approach using Convolutional Neural Networks (CNN) to detect pneumothorax in thoracic radiography images. The CNN model is trained on an annotated dataset with preprocessing steps, including zooming, brightness adjustment, flipping and format adjustment, followed by performance evaluation using accuracy, precision, recall, and F1 score metrics. Result: The results showed that the CNN model detected pneumothorax with 79.59% accuracy, a loss of 1.3056, and 1,092 correct predictions out of 1,372 test data. Precision was 51.12%, recall 78.62%, and F1 score 61.96%, confirming the system's potential, though further optimization is needed. Novelty: The novelty of this research lies in developing an automated pneumothorax detection system using a CNN architecture, improving diagnostic accuracy and efficiency. Despite high accuracy, precision and recall can be improved. Future research can focus on optimizing the model and applying data augmentation techniques.

Optimizing Image Feature Extraction with Convolutional Neural Networks for Chicken Meat Detection Applications

The food industry continuously prioritizes methods and technologies to ensure product quality and safety. Traditional approaches, which rely on conventional algorithms that utilize predefined features, have exhibited limitations in representing the intricate characteristics of food items. Recently, a significant shift has emerged with the introduction of convolutional neural networks (CNNs). These networks have emerged as powerful and versatile tools for feature extraction, standing out as a preferred choice in the field of deep learning. The main objective of this study is to evaluate the effectiveness of convolutional neural networks (CNNs) when applied to the classification of chicken meat products by comparing different image preprocessing approaches. This study was carried out in three phases. In the first phase, the original images were used without applying traditional filters or color modifications, processing them solely with a CNN. In the second phase, color filters were applied to help separate the images based on their chromatic characteristics, while still using a CNN for processing. Finally, in the third phase, additional filters, such as Histogram of Oriented Gradients (HOG), Local Binary Pattern (LBP), and saliency, were incorporated to extract complementary features from the images, without discontinuing the use of a CNN for processing. Experimental images, sourced from the Pygsa Group databases, underwent preprocessing using these filters before being input into a CNN-based classification architecture. The results show that the developed models outperformed conventional methods, significantly improving the ability to differentiate between chicken meat types, such as yellow wing, white wing, yellow thigh, and white thigh, with the training accuracy reaching 100%. This highlights the potential of CNNs, especially when combined with advanced architectures, for efficient detection and analysis of complex food matrices. In conclusion, these techniques can be applied to food quality control and other detection and analysis domains.

Development and implementation of a deep learning-based smart evaluation system for animated films

Animation has a rich history of over a century, and scholarly interest in the critique of animated films has evolved alongside its development. With the advent of the Internet and advancements in computer technology, traditional methods of animated movie appreciation are increasingly challenged by new complexities. This paper addresses these challenges by presenting an advanced intelligent system for evaluating animated films, utilizing cutting-edge deep learning models. Our approach begins with a pre-trained deep learning model to classify image categories and extract scene characteristics, while a primary color extraction algorithm captures key color features. Building on this, we propose an innovative deep convolutional neural network (CNN) augmented with fully connected layers to integrate multimodal features, enhancing the accuracy of film evaluations. Furthermore, we introduce a novel deep learning framework designed to detect pixel regions in images that evoke emotional responses. By employing visualization techniques within the convolutional network, we identify emotionally significant regions and enhance these features during initial training, improving the system’s emotional sensitivity. Experimental results demonstrate that our system not only improves the accuracy and efficiency of animated film evaluations but also surpasses existing methods, offering a more nuanced, emotion-aware approach to animation critique. This work advances the field by providing a robust, automated tool for evaluating animated films, offering deeper insights into their emotional and visual components, and contributing to the ongoing evolution of AI-driven film appreciation systems.

A guide to CNN-based dense segmentation of neuronal EM images.

Large-scale reconstitution of neuronal circuits from volumetric electron microscopy images is a remarkable research goal in neuroanatomy. However, the large-scale reconstruction is a result of automatic segmentation using convolutional neural networks (CNNs), which is still challenging for general researchers to perform. This review focuses on two representative CNNs for dense neuronal segmentation: flood-filling networks (FFN) and local shape descriptors (LSD)-predicting U-Net (LSD network). It outlines their basic mechanisms, requirements, and output segmentation using author's example segmentation. The FFN excels in segmenting long axons, and the LSD network is adept at segmenting myelinated axons. The choice between FFN and LSD depends on the target, as neither is universally superior. A common limitation of FFN and LSD is the easy detachment of thin spines from parent dendrites, which is fundamentally unavoidable. The author also introduces CNNs proposed to mitigate this issue. As CNN-based automated segmentation can take months, researchers need to be aware of the selection of an appropriate CNN, required computer resources, and fundamental limitations. This review serves as a guide for such dense neuronal segmentation.

Research on Concrete Beam Damage Detection Using Convolutional Neural Networks and Vibrations from ABAQUS Models and Computer Vision

Researchers have already used vibration data and deep learning methods, such as Convolutional Neural Networks (CNNs), to detect structural damage. Moreover, some researchers have employed image-based displacement sensors (such as the template matching and edge detection methods) to obtain structural vibration information. It is necessary to verify whether deep learning methods can detect minor damage inside beams, for example, small hollowing in concrete. In addition, there is an urgent need to develop an effective image-based displacement sensor that can simultaneously detect a large number of reliable vibration data from different measurement points. In this study, the vibration data of two beam-ABAQUS models were used as the input data for a newly designed deep learning-based structural health monitoring method. There were 500 vibration samples for each case, and the peak of vibrations was several millimeters. The proposed CNN model can locate damage positions in beams with high accuracy (close to 100%), and the damage sizes are 3 cm and 6 cm. Laboratory experiments were carried out on four beams with different damage. The optimized displacement sensor developed based on the edge detection method was used to detect the displacement of the beams. Each beam had 200 vibration data, and there were 800 vibration data in total. These vibration data were used as input data to train the proposed deep learning architecture, and satisfactory accuracy was achieved in detecting the damage of the beams with an accuracy of 97%. The training process is satisfactory in that the training loss and validation loss dropped very quickly.

A Novel Fault Diagnosis Method based on Convolutional Neural Network with Adaptive Noise Injection

Abstract In most existing intelligent fault diagnosis methods, noise is considered harmful and may decrease diagnosis accuracy. In contrast to these methods, this study proposes a novel fault diagnosis method with extra noise injection, termed an adaptive-noise-injected convolutional neural network (ANI-CNN). Noise is intentionally injected into a CNN model's softmax layer to improve fault diagnosis accuracy. The injected noise is used in the iteration of the CNN model and adaptively adjusted according to the change in model loss. Bearing datasets from Case Western Reserve University and Paderborn University were used to validate the effectiveness of the proposed method. The robustness of the proposed method was illustrated by injecting Gaussian and uniform noise. By comparing the ablation study results with those of the state-of-the-art methods, and t-test results before and after noise injection, the effectiveness of noise injection in enhancing diagnosis accuracy was demonstrated. The proposed method performed well on small samples and in complex working conditions, and possesses good generalizability and the ability to deal with real-world datasets.

Rapid Detection of Acetolactate Synthetase Inhibitor Resistant Weeds Utilizing Novel Full-Spectrum Imaging and a Hyperparameter-Tuned Convolutional Neural Network (CNN)

Abstract Herbicide-resistant weeds are fast becoming a substantial global problem, causing significant crop losses and food insecurity. Late detection of resistant weeds leads to increasing economic losses. Traditionally, genetic sequencing and herbicide dose-response studies are used to detect herbicide-resistant weeds, but these are expensive and slow processes. To address this problem, an artificial intelligence (AI)-based herbicide-resistant weed identifier program (HRIP) was developed to quickly and accurately distinguish acetolactate synthetase inhibitor (ALS)-resistant from -susceptible common chickweed plants. A regular camera was converted to capture light wavelengths from 300 to 1,100 nm. Full spectrum images from a two-year experiment were used to develop a hyperparameter-tuned convolutional neural network (CNN) model utilizing a “train from scratch” approach. This novel approach exploits the subtle differences in the spectral signature of ALS-resistant and -susceptible common chickweed plants as they react differently to the ALS herbicide treatments. The HRIP was able to identify ALS-resistant common chickweed as early as 72 hours after treatment at an accuracy of 88%. It has broad applicability due to its ability to distinguish ALS-resistant from -susceptible common chickweed plants regardless of the type of ALS herbicide or dose used. Utilizing tools such as the HRIP will allow farmers to make timely interventions to prevent the herbicide-escape plants from completing their life cycle and adding to the weed seedbank.

Real‐Time Nail‐Biting Detection on a Smartwatch Using Three CNN Models Pipeline

ABSTRACTNail‐biting (NB) or onychophagia is a compulsive disorder that affects millions of people in both children and adults. It has several health complications and negative social effects. Treatments include surgical interventions, pharmacological medications, or additionally, it can be treated using behavioral modification therapies that utilize positive reinforcement and periodical reminders. Although it is the least invasive, such therapies still depend on manual monitoring and tracking which limits their success. In this work, we propose a novel approach for automatic real‐time NB detection and alert on a smartwatch that does not require surgical intervention, medications, or manual habit monitoring. It addresses two key challenges: First, NB actions generate subtle motion patterns at the wrist that lead to a high false‐positives (FP) rate even when the hand is not on the face. Second, is the challenge to run power‐intensive applications on a power‐constrained edge device like a smartwatch. To overcome these challenges, our proposed approach implements a pipeline of three convolutional neural networks (CNN) models instead of a single model. The first two models are small and efficient, designed to detect face‐touch (FT) actions and hand movement away (MA) from the face. The third model is a larger and deeper CNN model dedicated to classifying hand actions on the face and detecting NB actions. This separation of tasks addresses the key challenges: decreasing FPs by ensuring NB model is activated only when the hand on the face, and optimizing power usage by ensuring the larger NB model runs only for short periods while the efficient FT model runs most of the time. In addition, this separation of tasks gives more freedom to design, configure, and optimize the three models based on each model task. Lastly, for training the main NB model, this work presents further optimizations including developing NB dataset from start through a dedicated data collection application, applying data augmentation, and utilizing several CNN optimization techniques during training. Results show that the model pipeline approach minimizes FPs significantly compared with the single model for NB detection while improving the overall efficiency.