Search

Due to the impact of the high-energy photons on the vision sensor, the images captured from the γ radiation environment are often polluted by the speckle noise with random location, disordered shapes, and uncertain color distribution. The uncertainty of the speckle noise leads to the lack of representation consistency of the texture features, brightness distribution, structural features, and other image descriptors for the radiation image, which seriously limits the feature expression of general feature descriptors for the key scene information. The limitation of representing the objects’ features in the radiation scenes directly aggravates the difficulty of the image alignment task. In this paper, we proposed a novel γ radiation scene image alignment approach jointed with a hierarchical noise suppression scheme and a sparse structure feature alignment strategy for the above-mentioned problems. The accuracy of γ image alignment can be boosted with three strategies, including removing the salient noise areas, suppressing the fine-texture noise regions, and the sparse representation of object features. Concretely, for the problem of object’s feature affected by the high saliency of the central region of the speckle noise in the radiation scene images, we firstly utilized the pyramid sliding sampling strategy to down-sample the γ radiation scene image into a set of lossless sub-images, where the speckle noise was divided into the isolated point noise in the sub-image. And then, a detection-based median-filtering scheme was presented to remove these isolated point noises. For the problem of insufficient saliency of feature representation for the scene objects in the image alignment task, a Sobel-based edge detection approach was used to extract the stable and sparse structure features of objects in the application scene, which can suppress the weak texture at the surrounding of the speckle noise and improve the robustness of scene object feature representation. Finally, the Fourier-Merlin transform, which is robust to different types of image noise, is utilized to accurately align the structural feature of the scene object to achieve the accurate alignment of the γ radiation scene images. In order to prove the effectiveness and generalization of our proposed method, extensive experiments were conducted on the γ radiation scene image set collected in the reality 60Co radiation environment. With the help of our proposed method, we can achieve the alignment accuracy that the scale analysis error is less than 0.01, the rotation analysis error is less than 04°, and the translation analysis error is less than ±06 pixel. Experimental results demonstrate the effectiveness and the generalization of our proposed method.

Pest migration has the characteristics of large scale and strong suddenness, which will lead to the outbreaks of pests and diseases, the decline of grain yield, and considerable economic losses. Entomological radar is an effective means of monitoring migratory pests. However, the Radar Cross Section (RCS) of an insect target is small, whereas the echo power is weak. High detection probability will result in a high false alarm probability. In the data association step of target tracking, the association error occurs due to the influence of false measurement. By utilizing the amplitude difference between the target and noise, the amplitude information-assisted tracking algorithm can effectively improve the recognition degree toward the target and noise and improve the tracking performance. However, the RCS fluctuation model of the target is needed as prior information to calculate the amplitude likelihood ratio. Therefore, in this paper, the insect RCS fluctuating characteristics are analyzed based on Ku-band entomological radar experiment data. The results show that gamma distribution can fit well the RCS probability distribution of the insect target. On this basis, we derive the amplitude likelihood ratio of the gamma fluctuation target in Gaussian white-noise background. By analyzing the simulation results and performance under different signal-to-noise ratios, measurement noises, and fluctuation model parameters, compared with probabilistic data association filter, the RCS feature-aided tracking algorithm can effectively improve the insect target tracking accuracy.