The Forward-Forward algorithm was developed to increase the resemblance of artificial neural network training processes to those occurring in the brain, in contrast to the backpropagation algorithm, which has been sho...
The Forward-Forward algorithm was developed to increase the resemblance of artificial neural network training processes to those occurring in the brain, in contrast to the backpropagation algorithm, which has been shown to have less similarity to brain processes. While Forward-Forward is a fascinating and novel idea, it significantly differs in performance from backpropagation. Forward-Forward strives to achieve this similarity by updating each layer independently of the others, through the introduction of a loss function that facilitates the separability of data from different classes. This inherent nature of creating discrimination between data of different classes inspired us to take advantages of contrastive learning to improve Forward-Forward performance. We modified the contrastive loss to be used in Forward-Forward, and our experimental results show that the proposed method improves the model accuracy and increases the convergence speed by more than 20 times.
Segmentation of brain tumor images is an important issue in medical imageprocessing and can help surgeons to accurately assess the tumor area. Since tumors vary in shape, size, and location from patient to patient, s...
Segmentation of brain tumor images is an important issue in medical imageprocessing and can help surgeons to accurately assess the tumor area. Since tumors vary in shape, size, and location from patient to patient, segmenting brain tumors is a challenge. In addition, small tumors are more difficult to segment than larger ones. In this paper, we present a method based on deep convolutional networks to improve the segmentation accuracy of brain tumors, especially small tumors in MRI images. In this method we have increased the accuracy of tumor segmentation by adding a module to UNet model. The proposed module uses deformable and dilated convolutions, which provide more spatial information to the network and thus increase the accuracy of tumor segmentation. The results show that our method is able to achieve a Dice of 0.8877 in the whole tumor section. For the core and enhancing tumor sections, we were able to achieve Dice values of 0.8683 and 0.8176, respectively.
This paper introduces a vision-based dynamic positioning (DP) control system and develops a hardware-in-the-loop (HIL) platform to validate the performance of the controller applied to a work-class remotely operated v...
This paper introduces a vision-based dynamic positioning (DP) control system and develops a hardware-in-the-loop (HIL) platform to validate the performance of the controller applied to a work-class remotely operated vehicle (ROV). The proposed platform consists of three main parts: hardware, imageprocessing part and controller. The hardware included a calibrated camera that was connected to a dedicated computer via USB 2.0. In the imageprocessing part after pre-processing a circular Hough transform was used to detect and determine the position of the target in the image plane. Furthermore, this paper proposed a feedforward proportional-integral-derivative (PID) controller. To evaluate the performance of the proposed controller, two scenarios were implemented. In the first scenario, the target was considered stationary and a disturbance was applied to the ROV in the simulation environment. In the second scenario, the target object has moved along a rectangular path, and the objective was to stabilize the ROV at the desired points. In both scenarios, the reference signal was acquired by a calibrated camera from the target and sent to the controller. The results showed the desirable performance of the proposed controller.
In the realm of agriculture and horticulture, machine vision and soft computing approaches have shown promise in overcoming the limitations of traditional methods for identifying plant illnesses utilizing various plan...
详细信息
On-road obstacle detection is an important field of research that falls in the scope of intelligent transportation infrastructure systems. The use of vision-based approaches results in an accurate and cost-effective s...
详细信息
Traffic accidents pose significant challenges to road safety and transportation management worldwide. Timely and accurate analysis of these incidents is crucial for effective response and mitigation. This paper presen...
详细信息
Breast cancer is the most commonly occurring cancer in women. Cancer patients frequently develop metastasis, which is responsible for more than 90% of their deaths. The mortality rate will be significantly reduced if ...
详细信息
In machine/computervision, cameras serve a major role in image acquisition. Surveillance scenarios typically rely on Closed-Circuit Television (CCTV) cameras. This study aims to evaluate industrial cameras within a s...
In machine/computervision, cameras serve a major role in image acquisition. Surveillance scenarios typically rely on Closed-Circuit Television (CCTV) cameras. This study aims to evaluate industrial cameras within a surveillance application, contrasting their performance with that of CCTV cameras. We explore the comparative analysis of CCTV and industrial cameras for vehicle attribute recognition, specifically concentrating on the recognition of vehicle color and model using deep learning techniques. To train and evaluate the models, we have created datasets from images captured by both a CCTV and an industrial camera. Our findings indicate that the industrial camera outperforms the CCTV. However, employing advanced processing algorithms has the potential to minimize the performance gap between these two cameras. Our research represents one of the initial comparative analyses between these camera types, offering valuable guidance in selecting the most suitable camera for specific applications.
Change detection in multi-temporal remote sensing data enables crucial urban analysis and environmental monitoring applications. However, complex factors like illumination variance and occlusion make robust automated ...
Change detection in multi-temporal remote sensing data enables crucial urban analysis and environmental monitoring applications. However, complex factors like illumination variance and occlusion make robust automated change interpretation challenging. We propose MaskChanger - a novel deep learning paradigm tailored for satellite image change detection. Our method adapts the segmentation-specialized Mask2Former architecture by incorporating Siamese networks to extract features separately from bi-temporal images, while retaining the original mask transformer decoder. To our knowledge, this is the first study in which change detection is converted from the existing per-pixel classification approach into a mask classification approach. Evaluated on the LEVIR-CD benchmark of over 600 very high-resolution image pairs exhibiting real-world rural and urban changes, MaskChanger achieves Fl-Score of 91.96%, outperforming prior transformer-based change detection approaches.
In mass disasters, quick identification of the victims is important even in rescuing the survivors. Teeth are one of the hard parts of the human body, which are more resistant to decay and destruction than other biome...
In mass disasters, quick identification of the victims is important even in rescuing the survivors. Teeth are one of the hard parts of the human body, which are more resistant to decay and destruction than other biometric modalities, and due to the variety of numbers, types, and shapes, they are considered a suitable option for identification. Due to the lack of Panoramic dental images from people before their death, the number of database images for training today’s modern identification systems is very limited which reduces the accuracy of recognition. In this article, three relatively time-robust features, the number of teeth, the number and position of restored teeth, and the interdental distances are used for identification. For the independence of identification to the variety of processes and the angle of photographs, we use spatial histograms of empty tooth areas as feature vectors. By comparing the Euclidean distance, the images of the data set can be recovered with a similarity of more than 85% to the input image, which is an acceptable accuracy for the simple and easy access of the features. Considering the expert’s role in the final decision, this system helps facilitate and accelerate identification.
暂无评论