Deep Learning-based Image Processing Research Articles

Recognition of car parts in automotive supply chains by combining synthetically generated training data with classical and deep learning based image processing

In the automotive industry, the ”completely knocked down” (CKD) business requires the identification and verification of a large number of unmarked parts. This is often a manual activity with a high risk of failure. Deep learning methods offer flexibility and a useful distinction between object classes for the automatic classification. However, these methods require large amounts of labeled training data and are sometimes inaccurate. In comparison, classical algorithms can achieve higher accuracy, but are inflexible and therefore unsuitable. This paper presents an approach to increase process reliability while alleviating the data labeling effort by using synthetic training data from spatially scanned and virtualized parts and combining deep learning and classical algorithms. The result is a flexible, sensor-independent, and autonomously trained classification system.

Remote Physical Frailty Monitoring-The Application of Deep Learning-Based Image Processing in Tele-Health.

Remote screening physical frailty (PF) may assist in triaging patients with chronic obstructive pulmonary disease (COPD) who are in clinical priorities to visit a clinical center for preventive care. Conventional PF assessment tools have however limited feasibility for remote patient monitoring applications. To improve the safety of PF assessment, we previously developed and validated a quick and safe PF screening tool called Frailty Meter (FM). FM works by quantifying weakness, slowness, rigidity, and exhaustion during a 20-second repetitive elbow flexion/extension task using a wrist-worn sensor and generates a frailty index (FI) ranging from zero to one; higher values indicate progressively greater severity of frailty. However, the use of wrist-sensor limits its applications in telemedicine and remote patient monitoring. In this study, we developed a sensor-less FM based on deep learning-based image processing, which can be easily integrated into mobile health and enables remote assessment of physical frailty. The sensor-less FM extracts kinematic features of the forearm motion from the video of 20-second elbow flexion and extension recorded by a tablet camera, and then calculates frailty phenotypes and FI. To test the validity of sensor-less FM, 11 COPD patients admitted to a Telehealth pulmonary rehabilitation clinic and 10 healthy young volunteers (controls) were recruited. All participants completed the test indicating high feasibility. Strong correlations (0.72 < r < 0.99) were observed between the sensor-based FM and sensor-less FM to extract all frailty phenotypes and FI. After adjusting with age and body mass index(BMI), sensor-less FM enables distinguishing COPD group from controls (p<0.050) with the largest effect sizes observed for weakness (Cohen’s effect size d=2.24), frailty index (d=1.70), and slowness (d=1.70). These pilot findings suggest feasibility and proof of concept validity of this sensor-less FM toward remote assessment of PF in COPD patients.

Deep Learning-Based Image Processing for Whitecaps on the Ocean Surface

Deep Learning-Based Image Processing for Whitecaps on the Ocean Surface

Deep Learning-based Image Processing Algorithms in 8K Era

Deep Learning-based Image Processing Algorithms in 8K Era

Deep Learning-based Image Processing Algorithms in 8K Era

Deep Learning-based Image Processing Algorithms in 8K Era

Multi-Modal Deep Guided Filtering for Comprehensible Medical Image Processing.

Deep learning-based image processing is capable of creating highly appealing results. However, it is still widely considered as a "blackbox" transformation. In medical imaging, this lack of comprehensibility of the results is a sensitive issue. The integration of known operators into the deep learning environment has proven to be advantageous for the comprehensibility and reliability of the computations. Consequently, we propose the use of the locally linear guided filter in combination with a learned guidance map for general purpose medical image processing. The output images are only processed by the guided filter while the guidance map can be trained to be task-optimal in an end-to-end fashion. We investigate the performance based on two popular tasks: image super resolution and denoising. The evaluation is conducted based on pairs of multi-modal magnetic resonance imaging and cross-modal computed tomography and magnetic resonance imaging datasets. For both tasks, the proposed approach is on par with state-of-the-art approaches. Additionally, we can show that the input image's content is almost unchanged after the processing which is not the case for conventional deep learning approaches. On top, the proposed pipeline offers increased robustness against degraded input as well as adversarial attacks.

Separation of Occluding Pigs using Deep Learning-based Image Processing Techniques

Separation of Occluding Pigs using Deep Learning-based Image Processing Techniques

From Signal to Image: Enabling Fine-Grained Gesture Recognition with Commercial Wi-Fi Devices.

Gesture recognition acts as a key enabler for user-friendly human-computer interfaces (HCI). To bridge the human-computer barrier, numerous efforts have been devoted to designing accurate fine-grained gesture recognition systems. Recent advances in wireless sensing hold promise for a ubiquitous, non-invasive and low-cost system with existing Wi-Fi infrastructures. In this paper, we propose DeepNum, which enables fine-grained finger gesture recognition with only a pair of commercial Wi-Fi devices. The key insight of DeepNum is to incorporate the quintessence of deep learning-based image processing so as to better depict the influence induced by subtle finger movements. In particular, we make multiple efforts to transfer sensitive Channel State Information (CSI) into depth radio images, including antenna selection, gesture segmentation and image construction, followed by noisy image purification using high-dimensional relations. To fulfill the restrictive size requirements of deep learning model, we propose a novel region-selection method to constrain the image size and select qualified regions with dominant color and texture features. Finally, a 7-layer Convolutional Neural Network (CNN) and SoftMax function are adopted to achieve automatic feature extraction and accurate gesture classification. Experimental results demonstrate the excellent performance of DeepNum, which recognizes 10 finger gestures with overall accuracy of 98% in three typical indoor scenarios.