Improved handling of motion blur for grape detection after deblurring
TL;DR: In this article, a deblurring-based method for detecting wine grapes was proposed. But, the deblur method was not applied to the detection of wine grapes, a crop with a variety of shapes, colors, sizes, and structures.
Abstract: Breakthroughs in the convolution neural network(CNN) have resolved and improved many challenges of pattern recognition in natural images. With the increased use of proximal sensing and low-cost cameras, monitoring and automation systems have gained popularity in the agriculture fields. Detection, segmentation, clustering, and counting are some fundamental problems associated with it. Here we are working on the detection of wine grapes, a crop with a variety of shapes, colors, sizes, and structures. Object detection is a challenging task especially when we are working on natural images. It is an even more difficult task when we are working on blurred images. Blur arises when images are taken via handheld camera, moving object in the automation system, or low rate video frame.Here we are trying to solve the motion blur problem in grape detection using three existing image deblurring algorithms. Performance of the deblurring algorithm is generally measured by peak-signal to noise ratio(PSNR) and structure similarity index(SSIM), but in addition to it, we have also considered blind/referenceless image spatial quality evaluator(BRISQUE). In this paper, we have comparatively analyzed: Scale recurrent network(SRN) for deep image deblurring, Multiscale convolution neural network for dynamic scale deblurring(Deep deblur), and DeblurGANv2: Deblurring(orders of magnitude) faster and better. Grape detection has experimented with yolov5x. Raw images from the standard dataset(GoPro and WGSID) were corrupted with various kinds of motion blurs. From the obtained result we can conclude that image deblurring significantly improves the performance of grape detection on the corrupted motion blur dataset.
Citations
More filters
TL;DR: In this article , a categorization of causes of image blur in precision agriculture is presented and a detailed introduction of general-purpose motion deblurring methods and their the strengthen and weakness is given.
TL;DR: In this paper , the performance of deep image denoising methods was evaluated in terms of improving the performance after image denoing, and it was shown that image dennoising does not improve the performance when applied to raw images of datasets.
Abstract: Image denoising is a process of inverse reconstruction where the original image is reconstructed from its noisy observations. Several deep learning models have been developed for image denoising. Usually, the performance of image denoising is measured by metrics like structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR), however in this paper, we take a more pragmatic approach. We design and conduct experiments to evaluate the performance of deep image denoising methods in terms of improving the performance of some popular computer vision (CV) algorithms after image denoising. In this paper, we have comparatively analyzed: fast and flexible denoising (FFDNet) convolution neural network (CNN), feed forward denoising CNN (DnCNN), and deep image prior (DIP)-based image denoising. CV algorithms experimented with are face detection, face recognition, and object detection. Standard and augmented datasets were used in our experiments. Various types and amounts of noise were added to raw images from standard datasets (BSDS500, LFW, FDDB, and WGSID). We may conclude from our findings that image denoising does not improve the performance of CV algorithms when applied to raw images of datasets. But image denoising is very effective in improving the performance of the CV methods when denoising is applied to noise corrupted images of the datasets. In our experiments, we found results where the improvements were up to 11.70% in terms of accuracy for the face detection experiment.
References
More filters
Proceedings Article•
01 Jan 2015TL;DR: This work introduces Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments, and provides a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework.
Abstract: We introduce Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments. The method is straightforward to implement, is computationally efficient, has little memory requirements, is invariant to diagonal rescaling of the gradients, and is well suited for problems that are large in terms of data and/or parameters. The method is also appropriate for non-stationary objectives and problems with very noisy and/or sparse gradients. The hyper-parameters have intuitive interpretations and typically require little tuning. Some connections to related algorithms, on which Adam was inspired, are discussed. We also analyze the theoretical convergence properties of the algorithm and provide a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework. Empirical results demonstrate that Adam works well in practice and compares favorably to other stochastic optimization methods. Finally, we discuss AdaMax, a variant of Adam based on the infinity norm.
111,197 citations
Posted Content•
TL;DR: The authors present some updates to YOLO!
Abstract: We present some updates to YOLO! We made a bunch of little design changes to make it better. We also trained this new network that's pretty swell. It's a little bigger than last time but more accurate. It's still fast though, don't worry. At 320x320 YOLOv3 runs in 22 ms at 28.2 mAP, as accurate as SSD but three times faster. When we look at the old .5 IOU mAP detection metric YOLOv3 is quite good. It achieves 57.9 mAP@50 in 51 ms on a Titan X, compared to 57.5 mAP@50 in 198 ms by RetinaNet, similar performance but 3.8x faster. As always, all the code is online at this https URL
12,770 citations
01 Jan 2012
TL;DR: The objective evaluation uses automatic algorithms to assess the quality of the image without human interfere.
Abstract: HE quality of the image degrades from the minute it is captured to the time it is displayed to the human observer. The image is subject to many kinds of distortions during the stages that it might pass through such as storing, processing, compressing, and transmitting, etc... In evaluating image quality there are two followed methods, the subjective and the objective method. The subjective method evaluation is considered costly, expensive, and time consuming; since we have to select a number of observers, show then a number of images and ask them to score images quality depending on their own opinion. The objective evaluation uses automatic algorithms to assess the quality of the image without human interfere.
199 citations
14 May 2017
TL;DR: The sparse regularization for the convolutional neural network (CNN) with the rectified linear units (ReLU) in the hidden layers is introduced and it is expected that the unnecessary increase of the outputs of the ReLU can be prevented.
Abstract: This paper introduces the sparse regularization for the convolutional neural network (CNN) with the rectified linear units (ReLU) in the hidden layers. By introducing the sparseness for the inputs of the ReLU, there is effect to push the inputs of the ReLU to zero in the learning process. Thus it is expected that the unnecessary increase of the outputs of the ReLU can be prevented. This is the similar effect with the Batch Normalization. Also the unnecessary negative values of the inputs of the ReLU can be reduced by introducing the sparseness. This can improve the generalization of the trained network. The relations between the proposed approach and the Batch Normalization or the modifications of the activation function such as Exponential Linear Unit (ELU) are also discussed. The effectiveness of the proposed method was confirmed through the detail experiments.
186 citations
23 Aug 2020
TL;DR: This work presents a large-scale dataset of real-world blurred images and ground truth sharp images for learning and benchmarking single image deblurring methods, and develops a postprocessing method to produce high-quality ground truth images.
Abstract: Numerous learning-based approaches to single image deblurring for camera and object motion blurs have recently been proposed. To generalize such approaches to real-world blurs, large datasets of real blurred images and their ground truth sharp images are essential. However, there are still no such datasets, thus all the existing approaches resort to synthetic ones, which leads to the failure of deblurring real-world images. In this work, we present a large-scale dataset of real-world blurred images and ground truth sharp images for learning and benchmarking single image deblurring methods. To collect our dataset, we build an image acquisition system to simultaneously capture geometrically aligned pairs of blurred and sharp images, and develop a postprocessing method to produce high-quality ground truth images. We analyze the effect of our postprocessing method and the performance of existing deblurring methods. Our analysis shows that our dataset significantly improves deblurring quality for real-world blurred images.
158 citations