Explore the vast range of titles available.

The identification of plant disease is an imperative part of crop monitoring systems. Computer vision and deep learning (DL) techniques have been proven to be state-of-the-art to address various agricultural problems. This research performed the complex tasks of localization and classification of the disease in plant leaves. In this regard, three DL meta-architectures including the Single Shot MultiBox Detector (SSD), Faster Region-based Convolutional Neural Network (RCNN), and Region-based Fully Convolutional Networks (RFCN) were applied by using the TensorFlow object detection framework. All the DL models were trained/tested on a controlled environment dataset to recognize the disease in plant species. Moreover, an improvement in the mean average precision of the best-obtained deep learning architecture was attempted through different state-of-the-art deep learning optimizers. The SSD model trained with an Adam optimizer exhibited the highest mean average precision (mAP) of 73.07%. The successful identification of 26 different types of defected and 12 types of healthy leaves in a single framework proved the novelty of the work. In the future, the proposed detection methodology can also be adopted for other agricultural applications. Moreover, the generated weights can be reused for future real-time detection of plant disease in a controlled/uncontrolled environment.

Early automation in identifying plant diseases is crucial for the precise protection of crops. Plant diseases pose substantial risks to agriculture-dependent nations, often leading to notable crop losses and financial challenges, particularly in developing countries. Symptoms such as chlorosis, structural deformities, and wilting, characterize these diseases. However, early identification can be challenging due to symptoms similarity. Researchers using artificial intelligence (AI) for plant disease classification, challenges like data imbalance, symptom variability, real-time performance, and costly annotation hinder accuracy and adoption. This work introduced a novel approach using the You Only Look Once (YOLO) deep learning model, chosen for its exceptional accuracy and speed. The study focuses on analyzing YOLO models, specifically YOLOv3 and YOLOv4, to identify fruit plant diseases. This work examines healthy peach and strawberry leaves, as well as peach leaves affected by bacterial spots and strawberry leaves with scorch disease. These models underwent thorough training using data from the publicly accessible Plant Village dataset. The simulation results were highly promising, numerically YOLOv3 model achieved 97% accuracy and a Mean Average Precision (mAP) of 92%, within a total detection time of 105 s. In comparison, the YOLOv4 model outperformed, with a 98% accuracy and an impressive mean average precision of 98%, all while completing the detection process in just 29 s. YOLOv4 demonstrated lower complexity, significantly faster, and more precise performance, especially in detecting multiple items. Serving as an efficient real-time detector, it holds the potential to transform plant disease diagnosis and mitigation strategies, ultimately leading to increased agricultural productivity and enhanced financial outcomes for developing nations.

YOLO (You Only Look Once) is a one-stage detector that predicts object classes and bounding boxes in a single pass without an explicit region proposal step. In contrast, two-stage detectors first generate candidate regions. The YOLO NAS-L model is specifically designed to improve the detection of small objects. The purpose of this study is to systematically investigate the influence of dataset characteristics, training strategies and hyperparameter selection on the performance of YOLO NAS-L in a challenging object detection scenario: detecting swimmers in aquatic environments. Using both the mean Average Precision value (mAP)—which reflects the model’s global precision–recall performance and the F1-score, indicating the model’s effectiveness under realistic operating conditions—as evaluation metrics, this study investigates the effects of batch size, batch accumulation, number of training epochs, image resolution, pre-trained weights, and data augmentation. Our findings indicate that while batch size and image resolution had limited impact on performance parameters, the use of batch accumulation, pre-trained weights and careful tuning of training epochs were critical for optimizing model performance. The results highlight the practical significance of combining optimized hyperparameters, training strategies, and pre-trained weights to efficiently develop high-performing YOLO NAS-L models.

Detection and calculation of wheat ears are critical for land management, yield estimation, and crop phenotype analysis. Most methods are based on superficial and color features extracted using machine learning. However, these methods cannot fulfill wheat ear detection and counting in the field due to the limitations of the generated features and their lack of robustness. Various detectors have been created to deal with this problem, but their accuracy and calculation precision still need to be improved. This research proposes a deep learning method using you only look once (YOLO), especially the YOLOv8 model with depth and channel width configuration, stochastic gradient descent (SGD) optimizer, structure modification, and convolution module along with hyperparameter tuning by transfer learning method. The results show that the model achieves a mean average precision (mAP) of 95.80%, precision of 99.90%, recall of 99.50%, and frame per second (FPS) of 22.08. The calculation performance of the wheat ear object achieved accurate performance with a coefficient of determination (R^2) value of 0.977, root mean square error (RMSE) of 2.765, and bias of 1.75.

Background In this article, we present a computerized system for the analysis and assessment of diabetic retinopathy (DR) based on retinal fundus photographs. DR is a chronic ophthalmic disease and a major reason for blindness in people with diabetes. Consistent examination and prompt diagnosis are the vital approaches to control DR. Methods With the aim of enhancing the reliability of DR diagnosis, we utilized the deep learning model called You Only Look Once V3 (YOLO V3) to recognize and classify DR from retinal images. The DR was classified into five major stages: normal, mild, moderate, severe, and proliferative. We evaluated the performance of the YOLO V3 algorithm based on color fundus images. Results We have achieved high precision and sensitivity on the train and test data for the DR classification and mean average precision (mAP) is calculated on DR lesion detection. Conclusions The results indicate that the suggested model distinguishes all phases of DR and performs better than existing models in terms of accuracy and implementation time.

Few-Shot Object Detection (FSOD) aims to identify new object categories with a limited amount of labeled data, which holds broad application prospects in real-life scenarios. Previous approaches usually ignore attention to critical information, which leads to the generation of low-quality prototypes and suboptimal performance in few-shot scenarios. To overcome the defect, an improved FSOD network is proposed in this paper, which mimics the human visual attention mechanism by emphasizing areas that are semantically important and rich in spatial information. Specifically, an Importance-Weighted Local Adaptive Prototype module is first introduced, which highlights key local features of support samples, and more expressive class prototypes are generated by assigning greater weights to salient regions so that generalization ability is effectively enhanced under few-shot settings. Secondly, an Imbalanced Diversity Sampling module is utilized to select diverse and challenging negative sample prototypes, which enhances inter-class separability and reduces confusion among visually similar categories. Moreover, a Weighted Non-Linear Fusion module is designed to integrate various forms of feature interaction. The contributions of the feature interactions are modulated by learnable importance weights, which improve the effect of feature fusion. Extensive experiments on PASCAL VOC and MS COCO benchmarks validate the effectiveness of our method. The experimental results reflect the fact that the mean average precision from our method is improved by 2.84% on the PASCAL VOC dataset compared with Fine-Grained Prototypes Distillation (FPD), and the AP from our method surpasses the recent FPD baseline by 0.8% and 1.8% on the MS COCO dataset, respectively.

Based on SCADA data, this study aims at fitting three performance curves (PCs), power curve, pitch angle curve, and rotor speed curve, to accurately describe the normal behaviour of a wind turbine (WT) for performance monitoring and identification of anomalous signals. The fitting accuracy can be undesirably affected by erroneous SCADA data. Hence, outliers generated from raw SCADA data should be removed to mitigate the prediction inaccuracy, so various outlier detection (OD) approaches are compared in terms of area under the curve (AUC) and mean average precision (mAP). Among them, a novel unsupervised SVM‐KNN model, integrated by support vector machine (SVM) and k nearest neighbour (KNN), is the optimum detector for PC refinements. Based on the refined data by the SVM‐KNN detector, several common nonparametric regressors have largely improved their prediction accuracies on pitch angle and rotor speed curves from roughly 86% and 90.6%, respectively, (raw data) to both 99% (refined data). Noticeably, under the SVM‐KNN refinement, the errors have been reduced by roughly five times and 10 times for pitch angle and rotor speed predictions, respectively. Ultimately, bootstrapped prediction interval is applied to conduct the uncertainty analysis of the optimal predictive regression model, reinforcing the performance monitoring and anomaly detection.

Object recognition is the technique of specifying the location of various objects in images or videos. There exist numerous algorithms for the recognition of objects such as R-CNN, Fast R-CNN, Faster R-CNN, HOG, R-FCN, SSD, SSP-net, SVM, CNN, YOLO, etc., based on the techniques of machine learning and deep learning. Although these models have been employed for various types of object detection applications, however, tiny object detection faces the challenge of low precision. It is essential to develop a lightweight and robust model for object detection that can detect tiny objects with high precision. In this study, we suggest an enhanced YOLOv2 (You Only Look Once version 2) algorithm for object detection, i.e., vehicle detection and recognition in surveillance videos. We modified the base network of the YOLOv2 by reducing the number of parameters and replacing it with DenseNet. We employed the DenseNet-201 technique for feature extraction in our improved model that extracts the most representative features from the images. Moreover, our proposed model is more compact due to the dense architecture of the base network. We utilized DenseNet-201 as a base network due to the direct connection among all layers, which helps to extract a valuable information from the very first layer and pass it to the final layer. The dataset gathered from the Kaggle and KITTI was used for the training of the proposed model, and we cross-validated the performance using MS COCO and Pascal VOC datasets. To assess the efficacy of the proposed model, we utilized extensive experimentation, which demonstrates that our algorithm beats existing vehicle detection approaches, with an average precision of 97.51%.

PurposeWith the rapid advancement of lifestyle and technology, human lives are becoming increasingly threatened. Accidents, exposure to dangerous substances and animal strikes are all possible threats. Human lives are increasingly being harmed as a result of attacks by wild animals. Further investigation into the cases reported revealed that such events can be detected early on. Techniques such as machine learning and deep learning will be used to solve this challenge. The upgraded VGG-16 model with deep learning-based detection is appropriate for such real-time applications because it overcomes the low accuracy and poor real-time performance of traditional detection methods and detects medium- and long-distance objects more accurately. Many organizations use various safety and security measures, particularly CCTV/video surveillance systems, to address physical security concerns. CCTV/video monitoring systems are quite good at visually detecting a range of attacks associated with suspicious behavior on the premises and in the workplace. Many have indeed begun to use automated systems such as video analytics solutions such as motion detection, object/perimeter detection, face recognition and artificial intelligence/machine learning, among others. Anomaly identification can be performed with the data collected from the CCTV cameras. The camera surveillance can generate enormous quantities of data, which is laborious and expensive to screen for the species of interest. Many cases have been recorded where wild animals enter public places, causing havoc and damaging lives and property. There are many cases where people have lost their lives to wild attacks. The conventional approach of sifting through images by eye can be expensive and risky. Therefore, an automated wild animal detection system is required to avoid these circumstances.Design/methodology/approachThe proposed system consists of a wild animal detection module, a classifier and an alarm module, for which video frames are fed as input and the output is prediction results. Frames extracted from videos are pre-processed and then delivered to the neural network classifier as filtered frames. The classifier module categorizes the identified animal into one of the several categories. An email or WhatsApp notice is issued to the appropriate authorities or users based on the classifier outcome.FindingsEvaluation metrics are used to assess the quality of a statistical or machine learning model. Any system will include a review of machine learning models or algorithms. A number of evaluation measures can be performed to put a model to the test. Among them are classification accuracy, logarithmic loss, confusion matrix and other metrics. The model must be evaluated using a range of evaluation metrics. This is because a model may perform well when one measurement from one evaluation metric is used but perform poorly when another measurement from another evaluation metric is used. We must utilize evaluation metrics to guarantee that the model is running correctly and optimally.Originality/valueThe output of conv5 3 will be of size 7*7*512 in the ImageNet VGG-16 in Figure 4, which operates on images of size 224*224*3. Therefore, the parameters of fc6 with a flattened input size of 7*7*512 and an output size of 4,096 are 4,096, 7*7*512. With reshaped parameters of dimensions 4,096*7*7*512, the comparable convolutional layer conv6 has a 7*7 kernel size and 4,096 output channels. The parameters of fc7 with an input size of 4,096 (i.e. the output size of fc6) and an output size of 4,096 are 4,096, 4,096. The input can be thought of as a one-of-a-kind image with 4,096 input channels. With reshaped parameters of dimensions 4,096*1*1*4,096, the comparable convolutional layer conv7 has a 1*1 kernel size and 4,096 output channels. It is clear that conv6 has 4,096 filters, each with dimensions 7*7*512, and conv7 has 4,096 filters, each with dimensions 1*1*4,096. These filters are numerous, large and computationally expensive. To remedy this, the authors opt to reduce both their number and the size of each filter by subsampling parameters from the converted convolutional layers. Conv6 will use 1,024 filters, each with dimensions 3*3*512. Therefore, the parameters are subsampled from 4,096*7*7*512 to 1,024*3*3*512. Conv7 will use 1,024 filters, each with dimensions 1*1*1,024. Therefore, the parameters are subsampled from 4,096*1*1*4,096 to 1,024*1*1*1,024.