RESUMEN
Deep learning based methods for medical images can be easily compromised by adversarial examples (AEs), posing a great security flaw in clinical decision-making. It has been discovered that conventional adversarial attacks like PGD which optimize the classification logits, are easy to distinguish in the feature space, resulting in accurate reactive defenses. To better understand this phenomenon and reassess the reliability of the reactive defenses for medical AEs, we thoroughly investigate the characteristic of conventional medical AEs. Specifically, we first theoretically prove that conventional adversarial attacks change the outputs by continuously optimizing vulnerable features in a fixed direction, thereby leading to outlier representations in the feature space. Then, a stress test is conducted to reveal the vulnerability of medical images, by comparing with natural images. Interestingly, this vulnerability is a double-edged sword, which can be exploited to hide AEs. We then propose a simple-yet-effective hierarchical feature constraint (HFC), a novel add-on to conventional white-box attacks, which assists to hide the adversarial feature in the target feature distribution. The proposed method is evaluated on three medical datasets, both 2D and 3D, with different modalities. The experimental results demonstrate the superiority of HFC, i.e., it bypasses an array of state-of-the-art adversarial medical AE detectors more efficiently than competing adaptive attacks1, which reveals the deficiencies of medical reactive defense and allows to develop more robust defenses in future.
RESUMEN
Changes in the canopy microclimate in intercropping systems, particularly in the light environment, have important effects on the physiological characteristics of photosynthesis and yield of crops. Although different row ratio configurations and strip widths of dwarf crops in intercropping systems have important effects on canopy microclimate, little information is available on the effects of intercropping on chlorophyll synthesis and photosynthetic physiological properties of dwarf crops. A 2-year field experiment was conducted in 2019 and 2020, with five treatments: sole maize (SM), sole peanut (SP), four rows of maize intercropping with eight rows of peanut (M4P8), four rows of maize intercropping with four rows of peanut (M4P4), and four rows of maize intercropping with two rows of peanut (M4P2). The results showed that the light transmittance [photosynthetically active radiation (PAR)], photosynthetic rate (Pn), transpiration rate (Tr), and stomatal conductance (Gs) of intercropped peanut canopy were reduced, while the intercellular carbon dioxide concentration (Ci) was increased, compared with SP. In particular, the M4P8 pattern Pn (2-year mean) was reduced by 5.68%, 5.33%, and 5.30%; Tr was reduced by 7.41%, 5.45%, and 5.95%; and Gs was reduced by 8.20%, 6.88%, and 6.46%; and Ci increased by 11.95%, 8.06%, and 9.61% compared to SP, at the flowering needle stage, pod stage, and maturity, respectively. M4P8 improves the content of chlorophyll synthesis precursor and conversion efficiency, which promotes the utilization efficiency of light energy. However, it was significantly reduced in M4P2 and M4P4 treatment. The dry matter accumulation and pod yield of peanut in M4P8 treatment decreased, but the proportion of dry matter distribution in the late growth period was more transferred to pods. The full pod number decreases as the peanut row ratio decreases and increases with year, but there is no significant difference between years. M4P8 has the highest yield and land use efficiency and can be used as a reference row ratio configuration for maize-peanut intercropping to obtain relatively high yield benefits.