This paper proposes a lightweight model combined with data augmentation for vehicle detection in an intelligent sensor system. Vehicle detection can be considered as a binary classification problem, vehicle or non-vehicle. Deep neural networks have shown high accuracy in audio classification, and convolution neural networks are widely used for audio feature extraction and audio classification. However, the performance of deep neural networks is highly dependent on the availability of large quantities of training data. Recordings such as tracked vehicles are limited, and data augmentation techniques can be applied to improve the overall detection accuracy. In our case, spectrogram augmentation is applied on the mel spectrogram before extracting the Mel-scale Frequency Cepstral Coefficients (MFCC) features to improve the robustness of the system. Then depthwise separable convolution is applied to the CNN network for model compression and migrated to the hardware platform of the intelligent sensor system. The proposed approach is evaluated on a dataset recorded in the field using intelligent sensor systems with microphones. The final frame-level accuracy achieved was 94.64% for the test recordings and 34% of the parameters were reduced after compression.

In radar target detection and tracking tasks, the detection algorithm and data association algorithm are the primary technologies. The accuracy of detection, stability of tracking and processing speed are the key points to accomplishing effective high‐resolution range profile (HRRP) multi‐target tracking (MTT). Classic HRRP target tracking algorithms conduct the detection by extracting handcraft features and conduct tracking based on the data association algorithm. With the development of neural network methods, deep learning methods have been widely applied in target detection and tracking. An HRRP detection and tracking (HDT) network based on neural networks is proposed, which includes three steps: Firstly, the HRRP signal is processed by a convolutional neural network detector to extract the feature of the origin signal and generate the detection measurement. Secondly, a predicted measurement is generated by a Kalman filter, which estimates the current state of the target based on its previous states and the detection measurement. Finally, the ReID network calculates the cosine similarity between the tracks and measurements to associate them and a linear sum assignment operation is used to match the tracks and measurements. By using the proposed algorithm, the HDT network is capable of detecting and tracking multiple targets in complex cluttered environments with high accuracy. Experimental results on an HRRP dataset collected by a moving radar platform show the outstanding performance of the HDT network, including detection accuracy, track integrity and real‐time MTT.