Green Citrus Detection and Counting in Orchards Based on YOLOv5-CS and AI Edge System

Green citrus detection in citrus orchards provides reliable support for production management chains, such as fruit thinning, sunburn prevention and yield estimation. In this paper, we proposed a lightweight object detection YOLOv5-CS (Citrus Sort) model to realize object detection and the accurate counting of green citrus in the natural environment. First, we employ image rotation codes to improve the generalization ability of the model. Second, in the backbone, a convolutional layer is replaced by a convolutional block attention module, and a detection layer is embedded to improve the detection accuracy of the little citrus. Third, both the loss function CIoU (Complete Intersection over Union) and cosine annealing algorithm are used to get the better training effect of the model. Finally, our model is migrated and deployed to the AI (Artificial Intelligence) edge system. Furthermore, we apply the scene segmentation method using the “virtual region” to achieve accurate counting of the green citrus, thereby forming an embedded system of green citrus counting by edge computing. The results show that the [email protected] of the YOLOv5-CS model for green citrus was 98.23%, and the recall is 97.66%. The inference speed of YOLOv5-CS detecting a picture on the server is 0.017 s, and the inference speed on Nvidia Jetson Xavier NX is 0.037 s. The detection and counting frame rate of the AI edge system-side counting system is 28 FPS, which meets the counting requirements of green citrus.

Underwater Fish Detection and Counting Using Mask Regional Convolutional Neural Network

Fish production has become a roadblock to the development of fish farming, and one of the issues encountered throughout the hatching process is the counting procedure. Previous research has mainly depended on the use of non-machine learning-based and machine learning-based counting methods and so was unable to provide precise results. In this work, we used a robotic eye camera to capture shrimp photos on a shrimp farm to train the model. The image data were classified into three categories based on the density of shrimps: low density, medium density, and high density. We used the parameter calibration strategy to discover the appropriate parameters and provided an improved Mask Regional Convolutional Neural Network (Mask R-CNN) model. As a result, the enhanced Mask R-CNN model can reach an accuracy rate of up to 97.48%.

Digital E. coli Counter: A Microfluidics and Computer Vision-Based DNAzyme Method for the Isolation and Specific Detection of E. coli from Water Samples

Biological water contamination detection-based assays are essential to test water quality; however, these assays are prone to false-positive results and inaccuracies, are time-consuming, and use complicated procedures to test large water samples. Herein, we show a simple detection and counting method for E. coli in the water samples involving a combination of DNAzyme sensor, microfluidics, and computer vision strategies. We first isolated E. coli into individual droplets containing a DNAzyme mixture using droplet microfluidics. Upon bacterial cell lysis by heating, the DNAzyme mixture reacted with a particular substrate present in the crude intracellular material (CIM) of E. coli. This event triggers the dissociation of the fluorophore-quencher pair present in the DNAzyme mixture leading to a fluorescence signal, indicating the presence of E. coli in the droplets. We developed an algorithm using computer vision to analyze the fluorescent droplets containing E. coli in the presence of non-fluorescent droplets. The algorithm can detect and count fluorescent droplets representing the number of E. coli present in the sample. Finally, we show that the developed method is highly specific to detect and count E. coli in the presence of other bacteria present in the water sample.

A Soft-YoloV4 for High-Performance Head Detection and Counting

Blockage of pedestrians will cause inaccurate people counting, and people’s heads are easily blocked by each other in crowded occasions. To reduce missed detections as much as possible and improve the capability of the detection model, this paper proposes a new people counting method, named Soft-YoloV4, by attenuating the score of adjacent detection frames to prevent the occurrence of missed detection. The proposed Soft-YoloV4 improves the accuracy of people counting and reduces the incorrect elimination of the detection frames when heads are blocked by each other. Compared with the state-of-the-art YoloV4, the AP value of the proposed head detection method is increased from 88.52 to 90.54%. The Soft-YoloV4 model has much higher robustness and a lower missed detection rate for head detection, and therefore it dramatically improves the accuracy of people counting.

Automating cell counting in fluorescent microscopy through deep learning with c-ResUnet

Counting cells in fluorescent microscopy is a tedious, time-consuming task that researchers have to accomplish to assess the effects of different experimental conditions on biological structures of interest. Although such objects are generally easy to identify, the process of manually annotating cells is sometimes subject to fatigue errors and suffers from arbitrariness due to the operator’s interpretation of the borderline cases. We propose a Deep Learning approach that exploits a fully-convolutional network in a binary segmentation fashion to localize the objects of interest. Counts are then retrieved as the number of detected items. Specifically, we introduce a Unet-like architecture, cell ResUnet (c-ResUnet), and compare its performance against 3 similar architectures. In addition, we evaluate through ablation studies the impact of two design choices, (i) artifacts oversampling and (ii) weight maps that penalize the errors on cells boundaries increasingly with overcrowding. In summary, the c-ResUnet outperforms the competitors with respect to both detection and counting metrics (respectively, $$F_1$$ F 1 score = 0.81 and MAE = 3.09). Also, the introduction of weight maps contribute to enhance performances, especially in presence of clumping cells, artifacts and confounding biological structures. Posterior qualitative assessment by domain experts corroborates previous results, suggesting human-level performance inasmuch even erroneous predictions seem to fall within the limits of operator interpretation. Finally, we release the pre-trained model and the annotated dataset to foster research in this and related fields.

Method of Step Detection and Counting Based on Measurements of Magnetic Field Variations

Basic human activity recognition (HAR) and analysis is becoming a key aspect of tracking and identifying daily habits that can have a critical impact on healthy lifestyles by providing feedback on health status and warning of deterioration. However, current approaches for detecting basic activities such as movements or steps rely on solutions with multiple sensors which affect their size and power consumption. In this paper, we propose a novel method that uses only a single magnetic field sensor for basic step detection, unlike the well-known multisensory solutions. The approach presented here is based on real-time analysis of magnetic field sensor measurements to detect and count steps during a walking activity. The approach is implemented in a system that integrates a digital magnetic field sensor with software blocks: filter, steady state detector, extrema detector with classifier, and threshold comparator implemented in an embedded platform. Outdoor experiments with volunteers of different ages and genders walking at variable speeds showed that the proposed detection method achieves up to 98% accuracy in step detection. The obtained results show that a single magnetic field sensor can be used to detect steps, and in general offers the possibility of simplifying the current solutions by reducing the device dimensions, the cost of a system and its power consumption.

Noninvasive Passenger Detection Comparison Using Thermal Imager and IP Cameras

Many modern vehicles today are equipped with an on-board e-call system that can send information about the number of passengers in the event of an accident. However, in case of fire or other major danger in a road tunnel, it is very important for rescue services to know not only the number of passengers in a given vehicle that has an accident and called help via e-call but how many people are in the tunnel in total. This paper deals with the issue of passenger detection and counting using the TPH3008-S Thermal camera and the VIVOTEK IP7361 IP Cameras noninvasively, i.e., the cameras are placed outside the vehicle. These cameras have their limitations; therefore, we investigated how to improve conditions and how to make detection better for future work. The main goal of this article is to summarize the achieved results and possibilities of improvement of the proposed system by adding other sensors and systems that would improve the final score of passenger detection. The experimental results demonstrate that our approach has to be modified and we have to add additional sensors or change methods to achieve more promising results. The results, findings and conclusions might be later used in tunnels and highways and also be applied in telematics and lead to better, safer road transport and improvement of existing tunnel systems sustainability by utilizing resources in a smarter way.