Abstract:Abstract In recent years, the field of bionics has attracted the attention of numerous scholars. Some models combined with biological vision have achieved excellent performance in computer vision and image processing tasks. In this paper, we propose a new bio-inspired lightweight contour detection network (BLCDNet) by combining parallel processing mechanisms of bio-visual information with convolutional neural networks. The backbone network of BLCDNet simulates the parallel pathways of ganglion cell–lateral geniculate nucleus and primary visual cortex (V1) area, realizing parallel processing and step-by-step extraction of input information, effectively extracting local features and detailed features in images, and thus improving the overall performance of the model. In addition, we design a depth feature extraction module combining depth separable convolution and residual connection in the decoding network to integrate the output of the backbone network, which further improves the performance of the model. We conducted a large number of experiments on BSDS500 and NYUD datasets, and the experimental results show that the BLCDNet proposed in this paper achieves the best performance compared with traditional methods and previous biologically inspired contour detection methods. In addition, BLCDNet still outperforms some VGG-based contour detection methods without pre-training and with fewer parameters, and it is competitive among all of them. The research in this paper also provides a new idea for the combination of biological vision and convolutional neural networks.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to improve the performance of contour detection, especially in terms of lightweight models. Specifically, the author proposes a new bio - inspired lightweight contour detection network (BLCDNet), aiming to achieve efficient contour detection by combining biological visual mechanisms and convolutional neural networks (CNN). Traditional methods have problems such as difficulty in distinguishing backgrounds and textures and low accuracy and performance when extracting image contours, while existing deep - learning methods, although having better performance, usually have high complexity, a large number of parameters and a slow processing speed. Therefore, the method proposed in this paper aims to overcome these shortcomings and design a lightweight contour detection model with a simple structure, few parameters, fast calculation speed and excellent performance.
### Main problems and solutions
1. **Limitations of traditional methods**:
- Traditional edge - detection methods (such as Prewitt, Sobel and Canny) mainly extract edges by calculating local grayscale changes in images, but perform poorly in distinguishing important backgrounds and textures, resulting in low accuracy and performance of contour extraction.
- Bio - inspired contour - detection methods, by simulating the characteristics of specific regions or cells in the biological visual system, suppress backgrounds and textures to a certain extent, but still have some problems, such as relying only on a single mathematical function to simulate complex neuron interactions.
2. **Complexity of existing deep - learning methods**:
- Although deep - learning - based contour - detection methods (such as RCF, BDCN, etc.) have achieved good performance, these methods usually have high complexity, a large number of parameters and a slow processing speed, and consume a large amount of computing resources.
- Many researchers enhance model performance by introducing pre - trained ImageNet parameters, but this increases the consumption of computing resources.
3. **Requirement for lightweight models**:
- In order to reduce the consumption of computing resources and increase the processing speed of the model, some researchers begin to explore methods to achieve high - performance image contour extraction under the conditions of simple models, a small number of parameters, fast operation speed and low resource consumption.
- Although existing lightweight models have solved the above problems to a certain extent, their designs mainly rely on the experience of researchers and lack the guidance of relevant biological visual mechanisms.
### Innovations of BLCDNet
To solve the above problems, this paper proposes BLCDNet, and its main innovations include:
1. **Simulating parallel pathways in the biological visual system**:
- A backbone network with three parallel channels is designed, which respectively simulates three parallel pathways from retinal ganglion cells to the lateral geniculate nucleus (LGN) and then to the primary visual cortex (V1).
- These channels include a large - receptive - field network (simulating the magnocellular pathway), a small - receptive - field network (simulating the parvocellular pathway) and a hybrid network (simulating the color pathway), and comprehensively extract and fuse feature information by fusing the outputs of these three channels.
2. **Deep feature extraction module**:
- A deep feature extraction module (DFEM) is designed using depth - separable convolution to further process the output of the backbone network, fully integrate context information and improve the overall detection performance of the model.
3. **Efficient lightweight architecture**:
- By combining the backbone network simulating parallel pathways and the designed deep feature extraction module, a simple - structured, efficient and accurate bio - inspired lightweight contour detection network (BLCDNet) is proposed. Experimental results on the BSDS500 and NYUD datasets show that this method is superior to traditional methods and other lightweight models in performance.
### Summary
The main contribution of this paper is to propose a lightweight contour detection network BLCDNet that combines biological visual mechanisms and deep - learning techniques. By simulating parallel pathways in the biological visual system and using depth - separable convolution, efficient and accurate contour detection is achieved. This method not only has made significant progress in performance, but also provides new ideas for future research.