Therefore dimensionless expression of the sensitivity is preferable,Bs(S0)=S0A��(S0)��dA��(S0)dS0,(19)where BS stands for the dimensionless sensitivity of the biosensor, A��(S0) is the steady state absorbance calculated at the substrate concentration S0 in bulk solution.We consider the dimensionless Biot number Bi to express the ratio of internal mass transfer resistance to the external one [34],Bi=d/DSe��/DSb=dDSb��DSe.(20)3.?Digital SimulationBecause
With the development of inexpensive multimedia hardware, such as micro-cameras and microphones, wireless multimedia sensor networks (WMSN) have recently emerged as an important technology, which has outstanding performance in multimedia signal acquisition and processing.

Compared to a wireless sensor network (WSN), WMSNs can not only enhance existing sensor network applications such as target tracking, classification, home automation, and environmental monitoring, but also enable several new applications, such as multimedia surveillance sensor networks, advanced health care delivery, industrial process control, and so on [1]. The content-rich vision-based information brings more effective way for target tracking and classification, but it also requires efficient distributed processing because the energy and network resources are strictly constrained in WMSNs. Target classification is a big challenge addressed in WMSN. The specific requirements of WMSNs, such as transmission of large amounts of data, various noises and time-varying samples, determine that the energy efficient, robust and online classifier learning algorithm is highly desired.

Support vector machine (SVM) is a well known classification tool, which has been widely used in WSNs [2-4]. AV-951 However, SVM classifier learning calls applies to solving a quadratic programming problem [5], which is computation expensive and cannot be afforded by single sensor node with limited computing ability and energy. Furthermore, traditional classifier learning is a kind of centralized learning strategy, which also needs to acquire samples from all sensor nodes. The large amount of data transmission will consume much energy, which is impractical for a highly constrained WMSN. Recently, various incremental learning methods were proposed [4-6], which implies that the learning process can be progressively carried out with the collaboration of multiple sensor nodes.

Although incremental learning is suitable for target classification in WMSNs, with the consideration of missing and false detection, the incremental learning should be collaboratively implemented according to the contribution of sensor nodes. This requires a new architecture to increase the system scalability with collaborative in-network processing by reducing the energy consumption and gathering the samples from a proper set of sensor nodes.