Particle Identification System in Cloud Chamber Using YoloV5
This study provides information to create a particle identification system that can identify traces left by particles with computer vision, based on the YOLO algorithm, using images captured by cloud chamber detectors.
Introduction
Many concepts in science are difficult to understand. Molecules, atoms, subatomic particles, and all of these have created an extremely useful framework of concepts in science. Although they are almost impossible to see and even more difficult to grasp mentally, many theoretical studies have been conducted that prove the existence of particles of all kinds. However, over time, strong empirical evidence for the existence of these particles has emerged. One of the biggest problems in physics has been the visualization of particles. The most revealing way to visualize them is to view them in diffusion cloud chambers. Even the smallest elementary particles trigger precipitation of liquid droplets, resulting in easily visible traces. Although cloud chambers have been replaced by much more technological devices today, they are unsurpassed when it comes to hands-on experience.
Real-time object recognition offered to us by the modern world is one of the leading topics of interest. Humans have visual and cognitive systems that can recognize objects by looking at an image. To recognize an object, systems such as computers take a classifier for that object and evaluate it at different locations and scales using test images. YOLOv5 is the fastest system that can perform this evaluation using current technology. YOLO (You Only Look Once) is amazingly simple. YOLO works with complete images and optimizes direct recognition performance. It is quite fast and runs the neural network on a new image at test time to predict recognition. For this reason, it is very important for real-time recognition.
Structure and Operation of the Cloud Chamber
While the observatory is being cooled, isopropyl alcohol is heated in the upper part (channel) of the interior of a cloud chamber. The isopropyl alcohol that evaporates from the channel sinks toward the cold observatory area. On its surface, it forms a supersaturated vapor layer about 1 cm thick.
When an electrically charged particle crosses the vapor layer of supersaturated isopropyl alcohol, its trail appears as a bright white line. The charged particle causes ionization of the isopropyl alcohol molecules, which act as a condensation nucleus. The observed white trace is a dense cloud of very small isopropyl alcohol droplets formed by rapid condensation.
A white trail formed after the flyby of a charged particle stands out from the black field of observation and is clearly visible even to the naked eye. No two cloud trails are alike, and they are different enough to determine which particle caused them and to determine their properties, sometimes even their origin. The formation of white trails is similar to the contrails produced by airplanes in the sky.
Observable Particles and Particle Interactions in the Cloud Chamber
As a result of the high energy of the particles, a trace is left along the trail of the charged particle caused by the many ions they produce. These traces are distinctive for particles. For example, an alpha particle’s trail is broad and flat, while an electron’s trail is thin and its deflection in collisions leaves a more pronounced trail. Different traces are evidence of the existence of different particles. There are various particles that are frequently observed, rarely observed, and very rarely observed within the cloud chamber.
OBJECT DETECTION USING TRANSFER LEARNING WITH YOLO V5
What is Computer Vision?
Computer Vision is an interdisciplinary scientific field that deals with gaining a high-level understanding of what the human visual system can do and automate it. The purpose of Computer Vision is to make computers see as much as humans as possible, that is, to give meaning to scenes or objects in life. In this context, giving meaning can be seen as decoding symbolic information from visual data. For computers, this is visual data such as photos and videos. Computer Vision Tasks can be listed as follows:
1. Recognition
The recognition task determines whether the image data contains an object or feature.
2. Contextual Image Acquisition
It allows to obtain the desired images by making distinctions from a wide set of images containing different types of content.
3. Exposure Estimation
It is used to estimate the position or direction of a determined object based on the camera shooting it.
4. Motion Analysis
5. Scene Rebuild
It is to provide multiple views of a scene or video to reconstruct the scene and calculate the 3D model.
Yolo Algorithm
Speed and accuracy are among the main features that make YOLO (you only look once), widely used and loved in the field of object detection, better than its competitors. In other object identification based algorithms, the image is first divided into approximately 2000 region recommendations and then CNN (ConvNet) is applied for each region respectively. The size of the regions is determined and the correct region is placed in the neural network. The biggest problem with this method is time. Since each region in the picture applies CNN separately, the training period takes quite a long time. In the YOLO algorithm, this process works differently.
The only difference between video and image processing is that images consist of a single frame, while videos consist of many frames. In images, the algorithm works for a single frame, while in videos it works repeatedly for all frames. The YOLO algorithm first divides the image into regions. It then draws the bounding boxes surrounding the objects in each region and calculates the probability of finding an object in each region.
It also calculates a confidence score for each bounding box. This score tells you with what percent probability the object is the predicted object. For example, if the confidence score for a found car is 0.3, it means that the probability of that object being a car is very low. In other words, YOLO says its prediction is reliable or unreliable.
If the confidence score is 0, there is no object, if it is 1, there is an object. More than one bounding box can be drawn for the same object within the same. To get rid of this problem, the non-maximum suppression technique I mentioned before is used. What is done with this technique is simply that the bounding box with the highest confidence score stays and the others are removed from view.
What is YOLOv5?
YOLOv5 is one of the latest object detection model. It is written in the framework of Ultralytics PyTorch, which is intuitive to use and provides very fast inferences. YOLOv5 is definitely easier to use and performs well on private data based on first runs. YOLOv5 has four different models, YOLOv5s, YOLOv5m, YOLOv5l and YOLOv5x, which offer different detection and performance power.
What is Google Colab?
With Google Colaboratory, it can develop deep learning applications using libraries such as Python, Keras, Tensorflow, PyTorch and OpenCV over Tesla K80 GPU etc. The most important feature of Google Colab is that it provides free GPU. GPUs are converters that work between the screen and the processor during the creation of text and graphics displayed on the computer. The GPU can process images fast thanks to its parallel processing architecture, which allows it to perform many calculations simultaneously. Thanks to the fast GPU system offered by Google Colab, models that would take days on ordinary computers can be completed in hours or even minutes. (Time may change depending on the application to be made.)
YoloV5 Particle Identification
In this study, a project that detects the particles observed in the cloud chamber will be prepared using Google Colab with the YOLOv5 algorithm. After this point, the steps to be taken will be explained step by step.
Data Generation and Labeling
First of all, the data set of the object to be detected is created. Depending on the object to be detected, 50–500 data will be sufficient for training.
A data annotation tool is needed to label the generated data. There are main programs to do this, such as labelImg and makesense. In this study, the makesense program was used for the labeling process. Each image in the dataset should be labeled. The important detail in the labeling process is that the process is done in accordance with the YOLO format. Both applications just mentioned have options suitable for the YOLO format. The tagging process is important so that the algorithm can distinguish the desired object and give it the train data that it can train itself.
In the .txt file shown in the figure above, the first row represents the “object_id”, that is, the number corresponding to the object category we listed. The second and third rows, center_x and center_y represent the center point of the bounding label. The fourth and fifth rows represent width and height.
Model Training Using Google Colab
After this stage, training of the YOLO model can be started via GoogleColab. GoogleColab’s official YOLOv5 program is available on its github site. The data set and tags created on the GoogleColab system are uploaded as zip files.
The COCO128.yaml file in the program is arranged according to the objects we want to detect and loaded into the program again. At this point, the important point is to use the files of the objects to be detected in the program.
After the loading process is complete, the cells are run. At this stage, the first step is taken to begin model training. While the model is being trained, not all of the data are included in the training at the same time. They take part in education in a certain number of parts. The first piece is trained, the performance of the model is tested, and the weights are updated according to the success with backpropagation. Then the model is retrained with the new training set and the weights are updated again. This process is repeated at each training step to try to calculate the most appropriate weight values for the model. Each of these training steps is called an “epoch”.
Since the most suitable weight values to solve the problem in deep learning are calculated step by step, the performance will be low in the first epochs, and the performance will increase as the number of epochs increases. However, after a certain step, the learning situation of our model will decrease considerably.
The model usually takes a long time to train; There are models that take days or months to train. This is common in deep learning. It has been epoched 5000 times for the Particle Identification System in the Cloud Chamber. This process took 1 hour even using GoogleColab.
Education Result
After this stage, the training phase of the object identification system has been completed. Test batch label files can be imported along with weight and width files from GoogleColab.
After the training is completed, the particle identification system can be applied over the desired image or video. Either the desired video or image can be processed via GoogleColab, or it can be processed via the generated python code. As a result, the particle identification system in the cloud chamber is now ready.
An Identification System Using YoloV5 and Images from Cloud Chamber Detectors
Pictures Taken on Video.
For YouTube Video: https://www.youtube.com/watch?v=7ShRrDpjNQk
Conclusion
Cloud Chambers is one of the oldest particle detectors and has led to many discoveries in particle physics. They are very important in terms of practical experience and mental comprehension.
In this study, the YOLO algorithm was used as the key element for particle detection with images taken from the Cloud Chamber, and it was seen that the model approach has an interesting potential and innovation. Particles lost in a short time in the Cloud Chamber were detected with a high confidence score, including the moments when the human visual system often fails to perceive. In this way, this system can be used for educational purposes by being used in real-time images taken from cloud chamber detectors in universities, and it can also be used as an educator in scientific centers. It also has applicability in many fields, especially in bubble chambers and other particle detectors. Real-time object detection with YOLOv5 has applicability in our daily life and other scientific fields, as well as particle detection in this study. For example, studies with the YOLO algorithm and other algorithms in the detection of cancerous cells in the field of medicine continue without slowing down.
Algorithms in this application can be developed with machine learning and/or numerical methods available in the scientific literature, and artificial neural networks can help in these studies.
Resources
[1] https://www.makesense.ai/
[2] https://tzutalin.github.io/labelImg/
[2]https://colab.research.google.com/github/ultralytics/yolov5/blob/master/tutorial.ipynb
[3] https://pytorch.org/hub/ultralytics_yolov5/
[4] https://github.com/ultralytics/yolov5
[5] https://colab.research.google.com/notebooks/intro.ipynb
[6] https://www.nuledo.com/en/
[7] https://roboflow.com/
[8] Hande Aygenli “Particle Identification System in CloudChamber Using YoloV5“ https://www.youtube.com/watch?v=7ShRrDpjNQk