Variable Receptive Field Research Articles

Multi-scale surface defect detection method for bottled products based on variable receptive fields and Gather–Distribute feature fusion mechanism

The inspection of visual quality represents a crucial step in the production process of bottled products. Numerous machine vision methodologies have demonstrated proficient identification of significant defects on bottle surfaces in well-controlled imaging environments. However, the actual production scenario introduces a myriad of surface defect types on bottled products, exhibiting diverse shapes, with the majority of defect instances characterized by relatively small individual areas. Faced with such diverse and small-sized defects, traditional convolutional neural networks (CNNs) encounter limitations due to the utilization of fixed-size convolutional kernels. This choice results in a constrained receptive field, hindering the capture of sufficiently effective contextual information. Additionally, pooling operations within traditional CNNs lead to a substantial reduction in feature map dimensions, causing excessive blurring or outright neglect of small-sized defects. Consequently, this detrimentally impacts the accuracy of surface defect detection and recognition. This study proposes a novel multi-scale defect detection model incorporating a variable receptive field and Gather–Distribute feature fusion mechanism to overcome limitations of traditional CNNs. Nine different scale defect images of bottled products, including worn, wrinkles, joint markings and etc, were collected and enhanced to construct a surface defect detection dataset for the actual production of bottled products. Enhancements to convolutional layers and the C2f module improve feature extraction for different defect shapes and sizes. Integration of the Gather–Distribute feature fusion module (GDFFM) reduces feature information loss and enhances shallow-layer feature utilization. An efficient detection head (E-Detect) with ”parameter sharing” is proposed to reduce computational complexity and improve detection speed without experiencing significant accuracy loss. Experimental results demonstrate that the model’s superiority over advanced defect detection algorithms across various categories. Notably, it achieves accuracies of 89.9%, 55.6%, 89.4%, and 75% for challenging defects like small worn, subtle wrinkles, inclined external cover, and misaligned labels, with improvements of 3.2%, 7%, 2%, and 11.8% over the baseline model, respectively. The model’s mean Average Precision (mAP) also increases by 2.7%.

Anchor-free object detection in remote sensing images using a variable receptive field network

Object detection is one of the essential tasks in computer vision, with most detection methods relying on a limited number of sizes for anchor boxes. However, the boundaries of particular composite objects, such as ports, highways, and golf courses, are ambiguous in remote sensing images, and therefore, it is challenging for the anchor-based method to accommodate the substantial size variation of the objects. In addition, the dense placement of anchor boxes imbalances the positive and negative samples, which affects the end-to-end architecture of deep learning methods. Hence, this paper proposes a single-stage object detection model named Xnet to address this issue. The proposed method designs a deformable convolution backbone network used in the feature extraction stage. Compared to the standard convolution, it adds learnable parameters for dynamically analyzing the boundary and offset of the receptive field, rendering the model more adaptable to size variations within the same class. Moreover, this paper presents a novel anchor-free detector that classifies objects in feature images point-by-point, without relying on anchor boxes. Several experiments on the large remote sensing dataset DIOR challenging Xnet against other popular methods demonstrate that our method attains the best performance, surpassing by 4.7% on the mAP (mean average precision) metric.

SLPR: A Deep Learning Based Chinese Ship License Plate Recognition Framework

Automatic ship license plate recognition (SLPR) for ship identification is of great significance to waterway shipping management. But few attention has been paid to SLPR in the past. In this paper, a novel cascaded Chinese SLPR framework consisting of the quadrangle-based ship license plate detection (QSLPD) algorithm and the rectification-based text recognition network (RTRNet) is developed. Concretely, in QSLPD algorithm, detection is performed based on the pyramid feature fusion architecture ameliorated by the proposed variable receptive field feature enhancement strategy and three task-specific output heads. In addition, a new loss function combining the dice coefficient and cross entropy is explored in the proposed SLPR which can generate significant improvement over the baseline. In RTRNet, regions of interest (RoIs) extraction and irregular text line rectification based on the vertices information predicted by QSLPD are performed before text recognition. Data augmentation are also applied to cope with the problem of limited text recognizer training data and the extremely imbalance distribution of corpus. Extensive experiments are carried out to demonstrate the reliability of the proposed cascaded SLPR framework, that can achieve the highest F-measure of 87.78% and 76.59% with IoU and TIoU metric on the collected dataset, surpasses many existing advanced methods.

Bidirectional Parallel Feature Pyramid Network for Object Detection

State-of-the-art Feature Pyramid Networks (FPNs) often focus on extracting features across different levels. In this paper, we propose a novel architecture, Bidirectional Parallel Feature Pyramid Network (BPFPN), to capture multi-scale spatial information from each level of FPN effectively. BPFPN consists of two blocks: Cross-level Channel Attention-Refinement (ClCSAR) Block and Weighted Parallel Feature Aggregation (WPFA) Block. ClCSAR block uses a channel attention mechanism to strengthen the context information of lower-level feature with aid from the upper-level feature. WPFA block exploits discriminating information from variable receptive fields via integrating multi-branch by employing dilated convolutions and using attention mechanisms to capture the salient dependencies over branches. Considering the incremental computation, we also give a lightweight version of BPFPN, namely BPFPN-Lite, integrated with an Efficient WPFA (E-WPFA) to improve detection accuracy while maintaining efficiency. Our proposed network can be easily plugged into existing object detection models and outperforms different feature pyramids methods by 0.2 ~ 2.1 on the COCO test-dev benchmark without bells and whistles.

CardioXNet: A Novel Lightweight Deep Learning Framework for Cardiovascular Disease Classification Using Heart Sound Recordings

The alarmingly high mortality rate and increasing global prevalence of cardiovascular diseases signify the crucial need for early detection schemes. Phonocardiogram (PCG) signals have been historically applied in this domain owing to its simplicity and cost-effectiveness. In this paper, we propose CardioXNet, a novel lightweight end-to-end CRNN architecture for automatic detection of five classes of cardiac auscultation namely normal, aortic stenosis, mitral stenosis, mitral regurgitation and mitral valve prolapse using raw PCG signal. The process has been automated by the involvement of two learning phases. Three parallel CNN pathways have been implemented in the representation learning phase to learn the coarse and fine-grained features from the PCG and to explore the salient features from variable receptive fields involving 2D-CNN based squeeze-expansion. Thus, in the representation learning phase, the network extracts efficient time-invariant features and converges with great rapidity. In the sequential residual learning phase, with the bidirectional-LSTMs and the skip connection, the network can proficiently extract temporal features without performing any feature extraction on the signal. The obtained results demonstrate that the proposed end-to-end architecture yields outstanding performance in all the evaluation metrics compared to the previous state-of-the-art methods with up to 99.60% accuracy, 99.56% precision, 99.52% recall and 99.68% F1- score on an average while being computationally comparable. This model outperforms the previous works using the same dataset by a considerable margin. The high accuracy metrics on both primary and secondary dataset combined with a significantly low number of parameters and end-to-end prediction approach makes the proposed network suitable for point of care CVD screening in low resource setups using memory constraint mobile devices.

HpGAT: High-Order Proximity Informed Graph Attention Network

Graph neural networks (GNNs) have recently made remarkable breakthroughs in the paradigm of learning with graph-structured data. However, most existing GNNs limit the receptive field of the node on each layer to its connected (one-hop) neighbors, which disregards the fact that large receptive field has been proven to be a critical factor in state-of-the-art neural networks. In this paper, we propose a novel approach to appropriately define a variable receptive field for GNNs by incorporating high-order proximity information extracted from the hierarchical topological structure of the input graph. Specifically, multiscale groups obtained from trainable hierarchical semi-nonnegative matrix factorization are used for adjusting the weights when aggregating one-hop neighbors. Integrated with the graph attention mechanism on attributes of neighboring nodes, the learnable parameters within the process of aggregation are optimized in an end-to-end manner. Extensive experiments show that the proposed method (hpGAT) outperforms state-of-the-art methods and demonstrate the importance of exploiting high-order proximity in handling noisy information of local neighborhood.

Rectangular-Normalized Superpixel Entropy Index for Image Quality Assessment.

Image quality assessment (IQA) is a fundamental problem in image processing that aims to measure the objective quality of a distorted image. Traditional full-reference (FR) IQA methods use fixed-size sliding windows to obtain structure information but ignore the variable spatial configuration information. In order to better measure the multi-scale objects, we propose a novel IQA method, named RSEI, based on the perspective of the variable receptive field and information entropy. First, we find that consistence relationship exists between the information fidelity and human visual of individuals. Thus, we reproduce the human visual system (HVS) to semantically divide the image into multiple patches via rectangular-normalized superpixel segmentation. Then the weights of each image patches are adaptively calculated via their information volume. We verify the effectiveness of RSEI by applying it to data from the TID2008 database and denoise algorithms. Experiments show that RSEI outperforms some state-of-the-art IQA algorithms, including visual information fidelity (VIF) and weighted average deep image quality measure (WaDIQaM).

Sensory noise predicts divisive reshaping of receptive fields.

In order to respond reliably to specific features of their environment, sensory neurons need to integrate multiple incoming noisy signals. Crucially, they also need to compete for the interpretation of those signals with other neurons representing similar features. The form that this competition should take depends critically on the noise corrupting these signals. In this study we show that for the type of noise commonly observed in sensory systems, whose variance scales with the mean signal, sensory neurons should selectively divide their input signals by their predictions, suppressing ambiguous cues while amplifying others. Any change in the stimulus context alters which inputs are suppressed, leading to a deep dynamic reshaping of neural receptive fields going far beyond simple surround suppression. Paradoxically, these highly variable receptive fields go alongside and are in fact required for an invariant representation of external sensory features. In addition to offering a normative account of context-dependent changes in sensory responses, perceptual inference in the presence of signal-dependent noise accounts for ubiquitous features of sensory neurons such as divisive normalization, gain control and contrast dependent temporal dynamics.

Pyramidal neural networks with evolved variable receptive fields

Pyramidal neural networks (PNN) are computational systems inspired by the concept of receptive fields from the human visual system. These neural networks are designed for implicit feature extraction and have been applied in pattern recognition applications. In the original approach, the size of the receptive field within the same 2D layer is a constant parameter, while in the human visual system, the receptive field size is variable. This paper proposes a PNN with variable receptive fields determined by an evolutionary algorithm, called variable pyramidal neural network with evolutionary algorithms. We observed from experiments aiming at detecting faces in images that our approach can achieve better classification rates than the original PNN. We also observed that regions with more information (such as nose and eyes) are more emphasized by variable receptive fields. These results confirm the application of intelligent algorithms to determine adjustable receptive fields in neural networks is useful to find out relevant information for recognition task. Besides, the model is comparable to biological systems regarding the flexibility assigned to receptive fields.

Vehicle Detection in Satellite Images by Hybrid Deep Convolutional Neural Networks

Detecting small objects such as vehicles in satellite images is a difficult problem. Many features (such as histogram of oriented gradient, local binary pattern, scale-invariant featuretransform, etc.) have been used to improve the performance of object detection, but mostly in simple environments such as those on roads. Kembhavi et al. proposed that no satisfactory accuracy has been achieved in complex environments such as the City of San Francisco. Deep convolutional neural networks (DNNs) can learn rich features from the training data automatically and has achieved state-of-the-art performance in many image classification databases. Though the DNN has shown robustness to distortion, it only extracts features of the same scale, and hence is insufficient to tolerate large-scale variance of object. In this letter, we present a hybrid DNN (HDNN), by dividing the maps of the last convolutional layer and the max-pooling layer of DNN into multiple blocks of variable receptive field sizes or max-pooling field sizes, to enable the HDNN to extract variable-scale features. Comparative experimental results indicate that our proposed HDNN significantly outperforms the traditional DNN on vehicle detection.

Stochastic Resonance in an Array of Locally-Coupled McCulloch-Pitts Neurons with Population Heterogeneity

SUMMARY We found a new class of stochastic resonance (SR) in a simple neural network that consists of i) photoreceptors generating nonuniform outputs for common inputs with random offsets, ii) an ensemble of noisy McCulloch-Pitts (MP) neurons each of which has random threshold values in the temporal domain, iii) local coupling connections between the photoreceptors and the MP neurons with variable receptive fields (RFs), iv) output cells, and v) local coupling connections between the MP neurons and the output cells. We calculated correlation values between the inputs and the outputs as a function of the RF size and intensities of the random components in photoreceptors and the MP neurons. We show the existence of “optimal noise intensities” of the MP neurons under the nonidentical photoreceptors and “nonzero optimal RF sizes,” which indicated that optimal correlation values of this SR model were determined by two critical parameters; noise intensities (well-known) and RF sizes as a new

A parallel analog intelligent vision sensor with a variable receptive field

The retina is an intelligent vision sensor indispensable in real-time image information processing of animals. The parallel image processing function of the retina was realized in an analog CMOS integrated circuit (vision chip). The chip was implemented with a one-dimensional Laplacian–Gaussian receptive field the size of which can be controlled by external signals. Active pixel sensors were used in the sensor part of the chip where the light sensitivity can be controlled by varying the accumulation time. Circuits compensating for noise due to the mismatch of transistor characteristics were incorporated, and the experimental results verified that sufficiently accurate outputs can be obtained under natural illumination. The chip can effectively extract the spatial frequency bandwidth of the signal in the presence of noise. © 1999 Scripta Technica, Syst Comp Jpn, 30(1): 60–69, 1999

A parallel analog intelligent vision sensor with a variable receptive field

The retina is an intelligent vision sensor indispensable in real-time image information processing of animals. The parallel image processing function of the retina was realized in an analog CMOS integrated circuit (vision chip). The chip was implemented with a one-dimensional Laplacian–Gaussian receptive field the size of which can be controlled by external signals. Active pixel sensors were used in the sensor part of the chip where the light sensitivity can be controlled by varying the accumulation time. Circuits compensating for noise due to the mismatch of transistor characteristics were incorporated, and the experimental results verified that sufficiently accurate outputs can be obtained under natural illumination. The chip can effectively extract the spatial frequency bandwidth of the signal in the presence of noise. © 1999 Scripta Technica, Syst Comp Jpn, 30(1): 60–69, 1999

Postnatal development of the nociceptive withdrawal reflexes in the rat: a behavioural and electromyographic study.

1. The postnatal development of nociceptive withdrawal reflexes was studied. In awake intact rats, forelimb, hindlimb and tail reflexes were recorded on videotape. In decerebrate spinal rats, electromyography (EMG) was used to record nociceptive withdrawal reflexes in musculi extensor digitorum longus (EDL), peronei, gastrocnemius-soleus (G-S) and biceps posterior-semitendinosus (BP-ST). Thermal (short-lasting CO2 laser pulses) and mechanical stimulation were used. 2. In adults, nociceptive withdrawal reflexes were typically well directed and reflex pathways to single hindlimb muscles had functionally adapted receptive fields. By contrast, at postnatal day (P) 1-7, the nociceptive withdrawal reflexes were often inappropriate, sometimes producing movements towards the stimulation, and EMG recordings revealed unadapted variable receptive fields. With increasing age, the nociceptive withdrawal reflexes progressively became well directed, thus producing localized withdrawal. Both withdrawal movements and spatial organization of the receptive fields were adult-like at P20-25. 3. Up to P25, reflex thresholds were more or less constant in both intact awake rats and spinal decerebrate rats, except in G-S in which no nociceptive withdrawal reflexes were evoked from P20 on. After P25, mechanical, but not thermal, thresholds increased dramatically. 4. EMG recordings revealed that during the first three postnatal weeks, the latency of the CO2 laser-evoked nociceptive withdrawal reflexes decreased significantly in peronei and BP-ST, but not in EDL, and thereafter increased significantly in peronei, BP-ST and EDL. The magnitude of the nociceptive withdrawal reflexes in these muscles increased markedly between P7 and P20 and showed little change thereafter. 5. Possible mechanisms underlying the postnatal tuning of the nociceptive withdrawal reflexes are discussed.