Multi-object garbage image detection algorithm based on SP-SSD

• Published in: Expert systems with applications Vol. 263; p. 125773
• Main Authors: Tan, Ling, Wu, Hui, Xu, Zifeng, Xia, Jingming
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 05.03.2025
• Subjects: ECM convolution module; Garbage detection; Residual connection; Spatial pyramid attention SPA module; Residual connection; ECM convolution module; Spatial pyramid attention SPA module; Garbage detection
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2024.125773

The existing garbage image detection methods are large in scale, long detection response time and low in accuracy. To solve these problems, this paper propose a multi-objective garbage image detection algorithm based on Self-Improvement Single Shot MultiBox Detector (SP-SSD). In the stage of feature extraction, depthwise separable convolution is used for downsampling, and a expansion convolution module (ECM) is designed to extract features. In the feature information fusion stage, residual connections are introduced for enhancing the generalization ability. In the stage of feature weight allocation, the spatial pyramid attention module (SPA) is introduced to enhance the representation ability. To verify the performance of the algorithm, this paper used 10,000 garbage images from the garbage dataset GCDD to carry out experiments. Experiments showed that compared with the classical SSD model, the mAP of SP-SSD is increased by 14.52%, parameters are reduced by 7 times, and the FPS is increased by 7. •Depthwise separable convolution is used for downsampling.•ECM module is used to extract features after high-low dimension conversion.•Residual connection is introduced to enhance the generalization ability.•SPA module is added to enhance the representation ability of the original features.