Sile Wang

Fast recognition of foreign fibers in cotton lint using machine vision

This paper presents an approach for fast segmentation of foreign fiber images and precise recognition of foreign fiber objects using machine vision. Live images were acquired in real time using a line scan CCD camera. After an image was acquired it was transferred to a host computer immediately for image processing and object classification. The captured image was firstly segmented according to the mean and standard deviation of R, G and B values of each pixel in the image. Then noises were removed using the area threshold method. Afterwards, color features, shape features and texture features of each foreign fiber object were extracted. Finally, a one-against-one directed acyclic graph multi-class support vector machine (OAO-DAG MSVM) was constructed and used to perform the classification. The results indicate that the image processing algorithm is fast and precise; the OAO-DAG MSVM gets a mean accuracy of 92.34% and a mean classification time of 12 ms, which can satisfy the accuracy and speed requirement of online classification of foreign fibers.

A New Approach for Greenness Identification from Maize Images

Greenness identification from crop growth monitoring images is the first and important step for crop growth status analysis. There are many methods to recognize the green crops from the images, and the visible spectral-index based methods are the most commonly used ones. But these methods can not work properly when dealing with images captured outdoors due to the high variability of illumination and the complex background elements. In this paper, a new approach for greenness identification from maize images is proposed. Firstly, the crop image was converted from RGB color space to HSV color space to obtain the hue and saturation value of each pixel in the image. Secondly, most of the background pixels were removed according to the hue value range of greenness. Then, the green crops were identified from the processed image using the excess green index method and the Otsu method. Finally, all noise objects were removed to get the real crops. The experimental results indicate that the proposed approach can recognized the green plants correctly from the maize images captured outdoors.

Saliency-based color image segmentation in foreign fiber detection

Abstract It is difficult to separate foreign fiber objects from the background in a live image captured by an automated visual inspection system for foreign fiber detection due to the inhomogeneous background brightness and various types of foreign fibers in different colors and shapes. This paper presents a saliency-based color image segmentation method aiming at foreign fiber detection. The RGB color image captured in real-time was firstly separated into R , G and B color channels. Then the red, green and blue color features were calculated respectively from the corresponding R , G and B channels. Afterwards, three saliency maps were obtained from these three color features and then fused together. The fused saliency map was segmented to get the color foreign fiber objects. Those foreign fiber objects in dark black or bright white were also segmented out using a threshold method from the brightness saliency map. Finally, all foreign fiber objects obtained were fused together to obtain the final objects. The results indicate that the proposed method can segment out color foreign fiber objects as well as gray foreign fiber objects in dark black or bright white.

A Review of Image Recognition with Deep Convolutional Neural Network

Image recognition technology is widely used in industry, space military, medicine and agriculture. At present, most of the image recognition methods use artificial feature extraction which is not only laborious, time consuming, but also difficult to do. Deep convolutional neural network is becoming a research hotspot in recent years. It has successfully applied to character recognition, face recognition, and so on. The traditional deep convolutional neural network still has some defaults when dealing with large-scale images and high-resolution complex images. So many research works are rolling ahead to improve the network to make it more efficient and robust. Firstly, the principle of the traditional convolutional neural network was briefly introduced. Then, the improvements on convolutional layer, pooling layer, activation function of convolutional neural network in recent years were summarized. Its applications in image recognition were also presented. Finally, the challenges in convolutional neural network research were analyzed and our recent works ware introduced.

Destriping Line-Scan Color Image in Transform Domain

To remove the stripe noises from the polluted line-scan color images, a transform domain destriping method which combined Fourier transform and wavelet decomposition was presented. Firstly, the line-scan color image with stripe noise was decomposed using multi-resolution wavelet to separate the sub-band which contained the stripe noise from the other sub-bands. Then the sub-band with stripe noises was transformed into Fourier coefficients. Thirdly, the Fourier coefficients were processed by a band-stop filter to remove these coefficients which represent the stripe noise. At last, inverse Fourier transform and wavelet reconstruction are performed to generate the final denoised image. The experimental results indicate that the proposed approach can remove the stripe noise effectively from the polluted image while preserving the details of the original image.

Fast Removal of Stripe Noise Based on Wavelet Decomposition

Vertical stripe noise, also called waterfall artifact, is generally occurred in the line scan images. It degrades the image quality and leads to object misrecognition. This paper presents a new approach for removal of vertical stripe noise using multi-resolution wavelet decomposition. The line scan image was firstly decomposed in highest L levels using wavelet decomposition. Then the vertical component in each decomposition level was discarded to remove the stripe noise. Finally, the destriped image was reconstructed from the L level processed components. The results indicate that the proposed approach can eliminate the stripe noise effectively from the polluted image.