Chassis Assembly Detection and Identification Based on Deep Learning Component Instance Segmentation

Abstract

Chassis assembly quality is a necessary step to improve product quality and yield. In recent years, with the continuous expansion of deep learning method, its application in product quality detection is increasingly extensive. The current limitations and shortcomings of existing quality detection methods and the feasibility of improving the deep learning method in quality detection are presented and discussed in this paper. According to the characteristics of numerous parts and complex types of chassis assembly components, a method for chassis assembly detection and identification based on deep learning component segmentation is proposed. In the proposed method, assembly quality detection is first performed using the Mask regional convolutional neural network component instance segmentation method, which reduces the influence of complex illumination conditions and background detection. Next, a standard dictionary of chassis assembly is built, which is connected with Mask R-CNN in a cascading way. The component mask is obtained through the detection result, and the component category and assembly quality information is extracted to realize chassis assembly detection and identification. To evaluate the proposed method, an industrial assembly chassis was used to create datasets, and the method is effective in limited data sets of industrial assembly chassis. The experimental results indicate that the accuracy of the proposed method can reach 93.7%. Overall, the deep learning method realizes complete automation of chassis assembly detection.

Fast semantic segmentation method for machine vision inspection based on a fewer-parameters atrous convolution neural network

“…Moreover, the speed problem is mainly based on practical applications. The atrous convolution architecture eliminates part of the CNN pooling layer while replacing the convolutional layer with a cascade or parallel atrous convolutional layer, enabling the analysis of the feature map at multiple arbitrary scales, and thus significantly improving the segmentation accuracy [11][12][13] and providing the possibility of detecting applications in the field of low power consumption. For obtaining a more accurate and faster fewer-parameters model as well as a method to achieve online machine vision and identification, in this study, the weight optimization technology https://doi.org/10.1371/journal.pone.0246093.g001…”

Semantic Segmentation under a Complex Background for Machine Vision Detection Based on Modified UPerNet with Component Analysis Modules

“…On the other hand, semantic segmentation can recognize the type of the object and divide the actual area at the pixel level, as well as implement certain machine vision detection functions, such as positioning and recognition [3]. As we start from image classification, move to object detection, and finally reach semantic segmentation, the accuracy of the output range and position information improves [4]. In the same manner, the recognition precision increases from the image-level to the pixel-level.…”

Transductive transfer learning for visual recognition

“…Image classification is an image-level visual recognition task that aims to classify each visual image into one of the pre-defined semantic categories; object detection is an instance-level visual recognition task that locates all the objects in a visual image and recognizes their semantic categories; semantic segmentation is a pixel-level visual recognition task that aims to assign a semantic category label to each and every pixel of an image. The progress in this research field enable a wide range of applications in computer vision, including autonomous vehicles [25][26][27][28][29][30], the analysis of medical images [31][32][33][34][35], the surveillance of manufacturing [37][38][39][40][41][42], construction [43][44][45][46], agriculture [47][48][49][50][51][52] and retail [53][54][55][56], and augmented and virtual reality in entertainment [57][58][59][60]. The technical methods of visual recognition can be broadly…”

“…object detection [21,22] and semantic segmentation [23,24]. In practice, visual recognition (i.e., classification, detection and segmentation) plays a significant role in various computer vision scenarios and applications including transportation (e.g., autonomous vehicles [25,26], drones [27,28] and robots [29,30]), healthcare (e.g., analysis of CT [31] and MRI [32,33] images, cancer detection [34,35] and patient movement analysis [36]), manufacturing (e.g., defect inspection [37,38], scene text recognition [39,40] and product assembly [41,42]), construction (e.g., predictive maintenance [43,44] and personal protective equipment detection [45,46]), agriculture (e.g., crop and livestock surveillance [47,48], automatic weeding [49,50] and insect detection [51,52]), retail (e.g., self-checkout [53,54] and surveillance for unmanned supermarkets [55,56]) and entertainment (e.g., augmented reality [57,58] and virtual reality [59,60]).…”