Dynamic Downscaling Segmentation for Noisy, Low-Contrast in Situ Underwater Plankton Images

Abstract

Finding and segmenting objects in noisy low-contrast in situ underwater plankton images is challenging because of the difficulty of separating potential plankton objects from the complex background and numerous and diverse other particles. In the present study, a dynamic downscaling model was developed to rapidly extract complete and clean regions of interest (ROIs) from images with highly variable content and quality. The original image was downscaled, and dynamic segmentation was performed in a scale pyramid space to ensure the integrity of weak targets based on local two-dimensional (2D) entropy parameters. Subsequently, a series of local thresholds and clustering gradients was examined iteratively for ROI selection. The performance of the local 2D entropy parameters relative to water turbidity (in a scattering medium and with high background noise) and image size was examined. To suppress the background and increase the sharpness of potential targets, a sharpness descriptor and gradient clustering were employed. The method was compared with the currently commonly used local threshold-based Sauvola segmentation using the same set of images. The results showed that the proposed method improves the ROI extraction accuracy and reduces oversegmentation for in situ underwater plankton images. It was concluded that the proposed method is a fast and robust segmentation technique and could facilitate the deployment of in situ plankton imaging systems for process-based research and routine plankton monitoring. INDEX TERMS dynamic downscaling, gradient clustering, in situ underwater plankton image, region of interest, scale pyramid space, segmentation.

Closed-Loop Region of Interest Enabling High Spatial and Temporal Resolutions in Object Detection and Tracking via Wireless Camera

“…The methods in these works typically capture a video frame, detect object(s) of interest with a neural network, then predict the ROI location of the next frame using an estimation filter such as a Kalman filter. Past applications include magnetic resonance imaging [10], lane boundaries detection for vehicle warning systems [11], in-situ plankton tracking [22], micro-object detection under a microscope [18], ultra-highspeed fruit fly wing tracking [12], ball and athlete detection for adaptive compression during sports streams [13], remote photoplethymography [23], hand prints and hand vein detection [24], [25], and blink detection to monitor patients with amyotrophic lateral sclerosis [26].…”

Development of a Buoy-Borne Underwater Imaging System for In Situ Mesoplankton Monitoring of Coastal Waters

“…Bright-field imagers support higher sampling throughput, so they were often towed by research vessels [16]- [20] or even carried by an autonomous glider [21] to capture images of plankton and report their underwater distribution. But they are usually poorer in imaging resolution and contrast [18], [19], and the images they acquired are susceptible to interference from the nonplankton particles especially in the coastal waters [22], [23]. Darkfield imagers generally have better resolution and contrast but less throughput, and they are more suitable for long-term high-frequency observations at fixed spots at the shore [24] or the sea floor [25].…”

Taming the data deluge: a novel end-to-end deep learning system for classifying marine biological and environmental images

“…Image segmentation is a long-standing problem in computer vision and most existing techniques are not suitable for noisy environments [15, 16]. Recent efforts on developing segmentation techniques, specifically for crowded underwater images, partially alleviate this issue [17, 18]; but given the complexity and uncertainty in underwater images, unsupervised deep learning approaches like region-based CNN (R-CNN) offer a more promising solution [19]. The R-CNN models generally include a region proposal network (RPN) to locate RoIs, a CNN model to describe features of RoIs generated from RPN proposals, and a classification layer to predict final bounding boxes and classes.…”